' Grassmannian Variables in Physics
by
Simon Edward Twisk, BSc(Hons)
in the Department of Physics submitted in fulfilment of the requirements for the degree of Doctor of Philosophy University of Tasmania October, 1988 Declaration I hereby declare that this thesis contains no material which has been accepted for the award of any other higher degree or graduate diploma in any tertiary institution and that, to the best of my knowledge and belief, this thesis contains no material previously published or written by any other person, except where due reference is made in the text of this thesis. ' Abstract
The five elements of this thesis are linked by the concept of Grassmannian variables. I begin with a brief introductory chapter discussing the general setting and then go on to deal with five topics each of which features anti-commuting co-ordinates in some guise. Chapter 2 uses the conventional, space-time supersymmetry, admissible for relativistic field theories. In the same way that the Dirac equation can be regarded as a square root of the Klein-Gordon equation, I have obtained a square root of the Dirac equation. This equation involves spinor-valued superfields and is given in terms of two component Grassmannian spinors. It has a larger component field content than the Dirac equation, just as the Dirac equation has a larger content than the Klein Gordon equation. After setting it up and solving the constraints for both the massless and massive cases, I have gone on for the massless case to solve the equation itself. The next chapter is concerned with Grassmannian variables in the context of path integrals. Specifically, I have studied the derivation of the index of the twisted Dirac operator via a supersymmetric quantum mechanics and taken great care to establish how certain ambiguities in the path integral can arise and how they can be circumvented. As an aside I have also obtained the general expression for the index of fields of arbitrary spin from the Atiyah-Singer index theorem itself. Chapter 4 uses anti-commuting co-ordinates as an appendage to the four commuting space-time co-ordinates. The Kaluza-Klein idea of force unification via general relativity is applied to a (4+N)-dimensional superspace. It is possible to give a consistent ansatz for a higher-dimensional metric which reproduces the standard model of elementary particles. I have considered the extension to grand unified theory and examined the SU(5) and SO(lO) models, showing how the former is more natural and more economical than the latter within such a framework. The consistell:~ quantization of a gauge theory requires the inclusion of "ghost" fields having "wrong" spin and statistics. The resulting gauge-fixed, quantized theory is endowed with a BRST symmetry, which replaces the classical gauge invariance. This symmetry can be best understood when considered with a partner, the anti-BRST symmetry. The two are both supersymmetries as they mix commuting and anti-commuting fields and can therefore be formulated on a superspace with two Grassmannian co-ordinates. In chapter 5, I sug,gest that it is useful to do this in an Sp(2)-symmetric manner - that is with the ghosts and anti-ghosts and the BRST and anti-BRST symmetries themselves treated symmetrically - but that extending the symmetry group to 0Sp(4/2) is more of a hindrance than a help. The final chapter of this thesis is concerned with a theory of massive, non abelian vector fields based on the Stueckelberg approach. Here renormalizability and unitarity are found to be conflicting requirements. Either one may be satisfied but not both. In particular, the violation of unitarity comes about either because of the failure of the BRST operator to be nilpotent or, diagrammatically, from the incomplete cancellation of the negative-norm ghost contributions. Acknowledgements I gratefully acknowledge the help and supervision of Prof.R.Delbourgo and of Dr.P.D.Jarvis. Also, I acknowledge the assistance of George Thompson and Ruibin von Zhang, with whom some of this work was done.
Q Table of Contents Title Page i Declaration ii Abstract iii Acknowledgements iv Table of Contents v Chapter 1 Introduction 1 Chapter2 Space-Time Supersymmetry and a Square Root of the Dirac Equation 1. Introduction 5 2. Superfields in N=l Supersymmetry 6 3. A Square Root of the Dirac Equation 10 4. Solution of the Constraints 13 5. Component Field Content 16 References 22 Chapter 3 Supersymmetric Quantum Mechanics and the Index Theorem 23 I. The Atiyah-Singer Index Theorem 1. Introduction 24 2. The Index Theorem for the Exterior Derivative of Differential Forms 26 3. The Index Theorem for the Dirac Operator 29 4. Gravitational Index Theorems for Arbitrary Spin Fields 30 5. The Index Theorem and Physics 34 II. Derivation of the Index Theorem via a Supersymmetric Quantum Mechanics 1. Preamble 37 2. Witten's Supersymmetry Index 37 3. Formulation of the Path Integral 39 4. The Index of the Dirac Operator 43 5. The Index of the Twisted Dirac Operator 48 References 56 Chapter 4 Grassmannian Kaluza-Klein Theory and Grand Unifi.cation 1. Preamble 58 2. Conventional Kaluza-Klein Theory 58 3. Supergravity 62 4. The Grassmannian Kaluza-Klein Ansatz 66 5. Matter Fields 69 6. Grand Unified Theories 71 References 74 Chapter 5 Sp(2)-BRST Quantization 1. Introduction 75 2. Faddeev-Popov Quantization and BRST 76 3. Sp(2)-BRST 81 4. Yang-Mills 83 5. Antisymmetric Tensor 86 6. Rarita-Schwinger 87 References 89 Chapter 6 Massive Yang-Mills Theory: Renonnalizability vs Unitarity 1. Preamble 90 2. Renormalizability of Massive Gauge Theories and the Stueckelberg Model 90 3. Unitarity 93 4. Renormalizability vs Unitarity 100 References 102 Conclusion 103 Appendix 105 Chapter 1 Introduction
The purpose of this chapter is just to give an outline of those areas in theoretical particle physics where Grassmannian or anticommuting variables occur, and then to mention the aspects which are relevant to each of the following chapters, where more detailed reviews will be given as appropriate. Canonical quantization of spinorial fields in relativistic quantum field theory leads to anticommutation relations for these fields and Fermi-Dirac statistics for the associated particles. This connection between spin and statistics within relativistic quantum field theory is perhaps its greatest triumph, explaining as it does why electrons obey the Pauli exclusion principle while photons do not. By contrast one loop effects are phenomenologically not as fundamental, though they do confirm the truth of quantum electrodynamics.
The h -7 co limit of the anticommutation relations leads to the consideration of totally anticommuting or Grassmannian fields. In the path integral approach to quantum field theory there is then the necessity of a definition of integration over such Grassmannian variables. This was supplied by Berezin - it is, for a Grassmannian variable 0, fd0 (0a + b) = a . In this thesis the convention for the ordering of multiple integrals will be such that 1 N N 1 Jd0 ... d0 0 ... 0 = 1 . The result for gaussian integrals which follows from Berezin's integration rule is t 1 0 0 2 JdN0 e- A oc: det A ' 1 rather than det-2A as for integration over commuting variables, and_this ensures agree- ment between the operator and the path integral formulations of quantum field theory. Anticommuting fields in violation of the spin-statistics theorem can still occur within the formalism of a relativistic quantum field theory provided they do not appear in the asymptotic states of the theory, that is provided they are not physical. Indeed such fields are, in general, necessary in the covariant treatment of theories which classically have a local gauge symmetry; these fields are the ghost fields. Of course, quantum mechanical models too can be considered with anticommuting fields. Then the strictures of the spin-statistics theorem no longer apply. A symmetry between commuting and anticommuting fields is called a supersymmetry. A consequence of the spin-statistics theorem is that the generators of any symmetry b,etween physical commuting and anticommuting fields in relativistic quantum field theory must themselves carry a spinor representation. At the same time, the Coleman-Mandula no-go theorem says that the largest Lie algebra of generators of a relativistic quantum field theory consists of Lorentz scalars, apart from Pµ and Mµv• the generators of translations and rotations. Thus the generators of a supersymmetry cannot reside within a Lie algebra. They may, however, be found within the odd part of a graded Lie algebra or superalgebra, that is they must satisfy anticommutation
1 relations amongst themselves. In this case they are still restricted so that they may only carry a spin 112 representation of the Lorentz group, that is they must be_ of the form Qx or Qa. The simplest supersymmetry algebra admissible is {Qa,~} =2~~Pµ
{Qa,Q~} = 0 = {0a~} (1)
[Pµ>Qal = 0 = [P w0al [Pµ>Pv] = 0 This may be extended by considering QxA, QxB, for A,B = l,... ,N, with A- _µ A {~,~B} =2cra~PµBB, or by admitting central charges so that {Qa,Q~} and {Q(t,Q~} are non-zero. We shall refer to this sort of supersymmetry as the conventional or space-time supersymmetry (space-time since it mixes with the space-time transfomations). It is' of course the supersymmetry of superstrings and supergravity. Quantum mechanical models can be viewed as (O+ 1)-dimensional quantum field theories. In this light,for such models the supersymmetry algebra (1) becomes {Q,Q} =2H, where His the hamiltonian operator of the model, with the other (anti)commutation relations vanishing. The remaining possibility for anticommuting fields which was mentioned above leads to another sort of supersymmetry. Between gauge fields and their attendant ghost fields there is the BRST supersymmetry. This symmetry is fundamental in that it ensures that the ghost fields do not appear in the outgoing asymptotic states, so that they do not threaten unitarity. Also, as the remnant within the covariantly quantized Yang-Mills theory of the classical, local gauge symmetry, it implies identities which are used to prove renormalizability. One way of constructing representations of a supersymmetry algebra is through a superfield construction on an appropriate superspace. A (d+N)-dimensional superspace is coordinatized by N Grassmannian variables, em, as well as d ordinary xµ. A superfield is a function on a superspace. Its dependence upon the Grassmannian coordinates is understood in terms of a power series with ordinary fields as coefficients, i.e. m F(~,8) = F(x) + 0 Xm(x) + ... , with the anticommuting nature of these coordinates ensuring that the series terminates at e1... eN. The component fields F(x),x;m(x), ... are each taken to be commuting or anticommuting in such a way that the superfield as a whole is one or the other. The superspace derivatives aµ = L and _a_ act on such superfields. axµ aem
2 Using
{a:m • a:m} = O , etc., they can be used to construct a representation of the supersymmetry algebra in the same way as position and orbital angular momentum can be represented in terms of aµ and xµ. Such a representation is then carried by the superfields and gives a representation of the algebra on the component fields through m m eQF(x,0) = SeF(x,0) = SeF(x) + 0 SeXm(x) + ... = eQF(x) + 0 eQxm(x) + ... In particular space-time supersymmetry can be represented on a (4+4)- dimensional superspace, (xµ,ea,eci) by ~ =-a--icf .e~µ, Qo. = ~ -i0ad\e~o.aµ a0a aa ae. a., a These operators obey {~.~} = - 2~a.Pµ, etc., with the change of sign here relative to (1) being necessary to ensure that the component fields carry a representation with the commutation relations (1). A final note on complex conjugation. As the order of writing Grassmannian quantities is important a convention for the effect of complex conjugation on ordering is required. Throughout this thesis we have taken complex conjugation as reversing the order of all Grasmannian quantities, in this way it is compatible with more general hermitian conjugation. This concludes our very brief review of the areas involving Grassmann variables. They are obviously very diverse and it is more appropriate to review the relevant aspects in each chapter as they arise; this is what we have done. The involvement of Grassmannian variables and/or supersymmetry in each of the following chapters is as follows. Chapter 2 uses the conventional space-time supersymmetry, and in particular the superspace differential operators to construct a sci,uare root of the Dirac equation involving spinor-valued superfields. It begins with a review of superfields in N=l supersymmetry. The next chapter deals with the Atiyah-Singer index theorem. There is a qerivation of the index for the twisted Dirac operator which uses path integrals within a supersymmetric quantum mechanics. We have examined this derivation, first locating the origin of certain ambiguities within the path integrals, and then circumventing them so as to obtain the correct final result. We found that the ambiguities do not have their origin in the Grassmannian integration rule. In the first part of this chapter, where we ~have reviewed the index theorem itself, we have also used it to derive the gravitational index theorems for fields of arbitrary spin. Chapter 4 involves the use of superspaces, not to describe a supersymmetry but rather to, in analogy to the Kaluza-Klein scheme, to attempt a unification of gravity and Yang-Mills theory within the framework of supergravity. For this purpose we
3 have reviewed the construction of supergravity. Specifically we have considered the grand unified models in this context. The last two chapters are concerned with ghost fields and the BRST supersymmetry. In Chapter 5 we have considered the formulation of the extended BRST supersymmetry on a (4+2)-dimensional superspace with an explicit Sp(2) symmetry of the Grassmannian coordinates maintained throughout. The conflicting nature of the requirements of renonnalizability and unitarity in amassive Yang-Mills theory without Higgs is the subject of Chapter 6. Here the failure of unitarity which we demonstrate is bound up with the failure of the generator of the BRST symmetry to be nilpotent or the inability of preventing the ghost fields from contributing to the outgoing asymptotic states. Work involved in Chapters 3,4,5, and 6 has been published in the following papers: P.D.Jarvis and S.Twisk, Class.Quantum Grav. 4 (1987). R.Delbourgo, S.Twisk and RB.Zhang, Mod.Phys.Lett. A3 (1988) 1073. R.Delbourgo, S.Twisk and G.Thompson, Int.J.Mod.Phys. A 3 (1988) 435. S.Twisk and R.B.Zhang, Mod.Phys.Lett. A3 (1988) 1169.
4 Chapter 2 Space-Time Supersymmetry and a Square Root of the Dirac Equation
1. Introduction Dirac initially found his equation by looking for a relativistic wave equation which would imply the Klein-Gordon equation, but which would be linear in atat like the Schrodinger equation, and hence, hopefully, yield a positive definite probability density. So, in a sense, the Dirac equation was found as a square root of the Klein Gordon equation. In so doing the argument of the equation became, instead of a scalar, a fou:r-component spinor - appropri;:i.te for the desciption of electrons and other spin-1'2 particles. If, in turn, we seek to find a square root of the Dirac equation, then we must find an operator whose square is i~ just as (@)2 = -D., The argument of this operator would presumably be from an enlarged space, perhaps yielding some physically interesting multiplet of fields. The spinorial differential operator D = (Da.,D ' - 1 13 - -~ - M+ We also give two square roots of the massive Dirac equation 5 0 A*X'I'(x,8~)J=rm('I'(x,8~)J, ( (4) A 0 B(x,8,8) B(x,8,8) which also involves the bosonic superfield B(x,8,S) in the codomain of A. Here the spinor-superfield 'I'(x,8,S) may be constructed in two different ways, either by (2) or by M+a =M 2. Superfields in N=l Supersymmetry Superfields are functions on superspace, F(x,8,S). They may take values in a space carrying some representation of the Lorentz group as well as of some internal symmetry group. Their 8,9 dependence should be understood in terms of their power series expansion into component fields, F(x,8, 9) = f(x) + 8a 6 to descibe an irreducible representation of the supersymmetry algebra a superfield must be constrained in some way. The operators a . _Jl _a. Da =-- + lCY .e aµ aea cm (7) n. = - _a_ - ieacf .a , a -&. aaµ ae satisfying {Da,Da.l = -2i~&.aµ_• anticommute with {Da,Qal = {Da,Oal = {Da.,Qal = {Da•Oal = 0 · Thus they can be used to place constraints on asuperfield in such a way that the constaint will be preseived under a supersymmetry transformation, i.e. in such a way that the constrained superfield will still carry a representation of the supersymmetry algebra. On a complex scalar-valued superfield S(x,9,S), the first such constraint is Da.S(x,0,S) = o . (8) Writing I = xµ + rncfe ' we have - µ - D.ya =0,D.0=0a giving the general solution to (8) S(x,9,9) = S(y,9) = A(y) + f2 0'Jf(y) + 90F(y) = A(x) + i0crµeaµA(x) - le 0000A(x) + f2 0'Jf(X) +-i e 0ScrµClµ'Jf(X) 4 ./2 + 99F(x). This multiplet of fields (A(x),'Jfa(x),F(x)) is known as the scalar or chiral multiplet and a superfield S(x,0,S) satisfting (8) as a scalar or chiral superfield, as is its conjugate S(x,9,9) which satisfies DaS(x,0,S) = 0 (9) The appropriate free supersymmetric action is 4 2 2 Id x Id 9 Id S (ss +; mSSo(S) +; mSSo(9)) = fd4x(-AOA + iClµ 'JfOµ 'I'+ FF+ m(AF + AF - ; 'Jf'I' - ; 'Jf'I')) This is the Wess-Zumino model. From (6) the supersymmetry transformations are o~A =fi ~'I' o~'I'= i./2 cf~aµA + 12 ~F o~F = i/2~crµaµ'I'. 7 The field F(x) acts only as a multiplier field and may be eliminated through its equation of motion giving the action 4 Jd x (-ADA- mZXA. + iaµ'l'O"µ 'I' - ; m(ff + 'l"I')), which is still invariant under a set of transformations mixing A and 'I'· However, this set of transformations will only close upon the supersymmetry algebra if the equations of motion for A and 'I' are used, that is the supersymmetry algebra only holds on shell. The other important multiplet of fields in flat supersymmetry can also be , formulated in terms of a constrained scalar-valued superfield. A scalar-valued superfield V(x,8,S) satisfying the reality constraint, V(x,8,8) = V(x,8,8) , (10) is known as a v:ector superfield. The appropriate free supersymmetric action is 4 2 2 fd xfd 8 d S (~ VDD2nv +m2VV). For m=O, this is invariant under the transformation v~v+ Vµ ~ Vµ + aµa , with a real. _a, (vµ(x), A.a.(x), A. (x), D(x)) is known as the vector multiplet. By allowing V to be Lie algebra valued this model may also be generalized to describe supersymmetric Yang-~ls theories. Other constraints involving higher orders of D, D may be imposed upon a scalar-valued superfield. While Da.S = 0, Da,S = 0 implies that S is independent of x, 2 2 D S = 0 , D S = 0 , S = S may be imposed. This yields the linear multipl.et (C(x), Xa.(x), Xa.(x), ~µ (x)) , with C and Aµ real and with aµAµ = 0. Such higher order constraints are not much discussed (in [2],[3] or [4] for example), perhaps because they tend to yield multiplets equivalent to the chiral or the vector multiplets and because they tend to give constraints upon the component fields which involve derivatives - such constraints would have to be put into the action by hand. 8 Superfields carrying overall non-trivial represntations of the Lorentz group are also not much discussed, except as regards supergravity. Constraints such as (8) Da.SCx,0,S) = o generalize immediately to other superfields as they do not mix with the overall representation. As sta~ed in the introduction, we will consider spinor-valued superfields with constraints of the form (aDaD a + bD.D--a. ) ,.-;-,.ti 1 -iJ... r w =--DDD v 4 for a real scalar-valued superfield V. A chiral spinor-valued superfield has also been considered in [5] and shown, with a certain choice of action, to be equivalent to the linear multiplet above. Supergravity, which I will discuss in Chapter 4, is based upon curved superspace admitting local supersymmetry transformations. The vielbein and connection generalize to superspace and are related by once again by a constraint on the torsion (not zero though, as in conventional gravity). In an appropriate supergauge the vielbein can be brought to 1- e~(x) ..L2 'l'cx(x) µ 2 'l'µa(X) A oa £M(x,O,O) = 0 m 0 0 0 ort: 9 3. A Square Root of the Dirac Equation In the chiral representation the Dirac equation is 0 icl. aµxcpa dx \jf (x)\jf (x) dx ( + f 1 2 =f Jdx efi(x)\jf2(x) which decomposes into the inner products fdx cpr(x)cp2ex(x) and fdx x 1 a(x)x~(x) on the spaces of positive and negative chirality spinors respectively - that is (i9)'.): = (i(/)_ . In the next chapter these facts (or their Euclidean versions) will be exploited to draw an analogy between the Dirac operator acting between the spaces of positive and negative chirality spinors and a supersymmetry operator acting between bosonic and fermionic spaces ip. order to find the index of the Dirac operator. Here we seek an operator, A, which acts on a spinor-valued superfield, 'P(x,0,S), of space-time supersymmetry and for which Since - ~ {Dex,Da} = (ia)+ 1 - . - 2 {Dex,Da} = (1a)_, such an A inight be linear in Dex, Da, . Then A 'I' (x,0,0) would be a bosonic superfield - that is a commuting tensorial superfield B(x,0,S). The square root of the Dirac equation would be 0 A*X'l'(x,0~)) = .fiii ('l'(x,0~)) ( (15) A 0 B(x,0,8) B(x,0,8) implying that i~(x,0,S) = m'l'(x,0,S) . Let 10 4 2 2 ('I' ,'l' ) = Jd x d 0 d S 'I' (x,0,S)'l' (x,0,S) (16) 1 2 1 2 be the inner product on the space of pirac spinor-valued supetfields, 2 2 (Sl'S ) = d\ d 0 d S (x,0,S)S (x,0,S) (17) 2 f S1 2 on scalar-valued superfields and 2 4 2 (18) (V1,V2) = fd x d 0 d SVC:a (19) 'has dual given by (A'l',B) = ('l',A*B) (20) where B= (~). Now on a superfield F(x,0,S) DaF=±DaF according as F is commuting or anti-commuting, and similarly ,DaF=±DaF, since the conjugation reverses the order of any factors as explained in the appendix. Also, since Da(F1F2) = (Dcl1)F2 ± F1(DaF2)' the rule for integration by parts is fd20 (DaF1)F2 = + fd20 Fl (DaF2) , once again according as F1 is bosonic ot fermionic. Similarly, for Da fd28 (DaFl)F2 = +f d28 Fl (DaF2) . Thus 4 2 2 (A'¥,B) = ;, Jd x d 0 d 0 ( (D .oI>" -D.x">s + (DP '11 ~ + D ~Xp)V: J 4 2 2 =-1 fd x d 8 d 0 (- (21) 11 For this choice of the operator A then where we have used the identities ~a~P = +~\.p~, ~a~~= --}~~"o:. p - -~ Writing M(a,b) = aD DP+ bD~D , we have -M(-,-)a1 1 + icr_Jl ilaµx-~) A* A'I! = 4 2 ap · ( ·cflaPa - M(1 l)Xa 1 µ ~ 2'4 =. -(M but this does not imply any further conditions. An alternative set of conditions to achieve a square root of the Dirac equation is 12 q M(.!. .!.) = M 4'2 a a (24) i 1-a _a M(--)X =MX . 2'4 Then A* A'I' = id'P - M'I' and the equation (15) implies itN' = (m+M)'I' . (25) We now look at the solutions of the constraints (23) and (24). 4. Solution .Qf ~ Constraints As the op(1rator M(a,b) = anan11 + bDaDa does not interact with the overall index of a superfield, we may as well suppress this index - provided we keep the component fields on the right so that their commutativity is not important. o"D .. = e"P (a:P +i~pe\) (a:"+ icr~,;a"a,,) 1113 = e Ll+ ieal3 (- CJV. eaavL+ cl:. a _a_)- eal3cf'. CJV • .!.eaPeea a ael3 aea aa ael3 1313 µaea 1313 aa 2 µ v all a a Jl a -Ct a c - =e P--+2i(ecr) .8 -a +880 (26) ae13 aea a aea µ and (27) Let the expansion in 8,0 of a supe~eld F be F(x,0,S) = f(x) + 8acpa(x) + Sa'Xa(x) + 00m(x) + 08n(x) + 8aµSvµ(x) + 888/..(x) + 888\j/(X) + 8888d(x) (28) Then a - -- D Da.F(x,8,8) = -4m(x) - 40/..(x) - 488d(x) _µ a- 13 - - µ - µ = - 4m(x) - 28(2A.(x) + rcr aµcp(x)) + 88(-4d(x) - 2ia vµ(x).+Of(x)) + 4iecfeaµm(x) + 888(0cp(x) - 2iaµaµr(x)) + eeeeom(x) (29) and - _a - D. D F(x,8,8) = - 4n(x) - 48'\jl(x) - 498d(x) a + 2i8a.(cfe):c-aµXa(x) - 2eaaµn(x) - ep a~aµvv(x)- iea0Paµ'l'p(x)) + 080f(x) + 0800 x(x) + '80880n(x) = - 4n(x) - 20(2'\jl(x) + icfaµx(x)) + 08(-4d(x) + 2iaµ vµ(x) +Of(x)) - 4i0cf8aµn(x) + 80S(OX(x) - 2icrµaµ'l'(x)) + 08000n(x) . (30) Thus M(a,b)F(x,e,e)· = MF(x,8,S) (31) implies that - 4am(x) - 4bn(x) = Mf(x) (31a) - 2b(2'1'(x) + icfaµxCx)) = Mcp(x) (3lb) - 2a(2r(x) + icrµaµcp(x)) = Mx;(x) (3 lc) b(-4d(x) + 2iaµvµ(x) +Of(x)) = Mm(x) (31d) a(-4d(x) - 2iaµvµ(x) +Of(x)) = Mn(x) (31e) - 4iaaµm(x) - 4ibaµn(x) = Mvµ(x) (31f) b(ffi(x) - 2icrµaµ'l'(x)) = W:(x) (3lg) a(Ocp(x) - 2iaµa~(x)) = M'\jl(x) (31h) aOm(x) + b0n(x) = Md(x) (31 i) We consider first the case that M = 0, a,b :;i!: 0. Then (31a) and (3 lf) imply (3 li) and that m(x) = c , a constant, and n(x) = ~c. (31b) '\jl(x) = _j_ cfa x(x) 2 µ implies (31g). (3lc) - . µ A.(x) = -.!.. a cp(x) 2 a µ implies (3lh). (31d) + (31e) imply that d(x) =!.Of(x) 4 and (3ld)- (3le) that 14 So the general solution to M(a,b)F(x,0,0) = 0 is F(x,0,S) = f(x) + 0 + ecte(~ aaµm(x) + id/(x)) + ~ eeecrµaµcp(x) - ~ 009cp(x) (35) M2 - + ab eee Sf(x) , 64 with - M2 - OF(x,9 ,9) =- ab F(x,9,0) . 16 The supersymmetry transformations of the component fields in both cases can 15 be found through (6) - a --a - B F(x,0,0) = (~ <2a + ~. Q )F(x,0 ,0) . ~ a For the multiplet of (32), (f(x), B~f(x) = ~cp(x) + ~x(x) B~ 1 B~X,(x) = - ~ ~m- cr~ ~(vµ(x) + i()µf(x)) (36) oc=O ~ B~vµ(x) = ~ ~(CJV~ _ cfcrv)ay B~ 8~m(x) = i~crµaµ 5. Component Field Content Using the results of the previous section, we can now write down the spinor valued superfields upon which the operator A is indeed the square root of the Dirac operator. For the massless case A 'I'(x,9,0) =- 1 (D:a - Dax:)= o (38) f2 D" + D X" 13 13 implies that iJ'I'(x,9,0) = o, if 13 a(x,9,S) = a(x) + 0 ~~a(x) +(eaµ E)avµ(x) + 99'1'a - ; 09'1'a + ecfeA.µa(x) . - _µ1313 . - ~v - - ~ 900PCJ aµm13a(x)- ~ 00(0CJ CJ e)aaµvv(x) + ! 00000cpa(x) (39) and 16 x<\x,e,0) = x«Cx) + cecfe)awµCx) + epn~'\x) + eero« __ 2eero« + ecle~:_Cx) - J... 88 (S"fl(JVE)ad W (x) - J... 800p (Jµ. dµnpa(x) + ..!.. 8 0SS0 x'\x) , (40) 2 µ v 2 pp 4 µ ":\µ~ -a . where a Aµa(x) = 0, o ~µ(x) = 0 and 'l'a, ro are constant spmors. For the massive case 0 A*X'P(x,0~)) = rm ('P(x,8~)) ( A 0 B(x,8,8) B(x,8,8) implies that :id"P'(x,e,0) =m'P(x,8,S) if 'P(x,8,8) is given by (38) and (39), or that i.d'P(x,8,S) = (m+M)'P(x,8,0) (41) if p . . 17 AA"'B(x,0,S) = mB(x,0,S) , (45) which is (46) a - -a M(l,l) = D Da + DaD . This equation still involves 0-derivatives, even if we impose M(l,l)B =MB. This is perhaps not unreasonable as a bosonic field should obey a second order equation involving m2. In any crise, we can ignore B(x,8,S) in consideration of the massless equation for 'l'(x,8,0), (38). With 'I' given by (39), ( 40) a - --a - D _ aP( _µ _p . - _µpp i - -~v -E mpa(x)+20p'lfa+rrpp8 A.µa(x)-10P8p0' aµmpa(x)- 2 00(rr O' E)paaµvv(x) - ) P a-a( p - µ p + ~ 8p000 -_µ i - _µpp ) 0apA.µ (x)- e00Pa aµapmpa(x) -ea 0 2 pa - a µ- · µpp a - P µ pci + £. n (X) + 48 • (J) - 00" ~µ(X) - .!..08e. (j (O'V£)A aµWy(X) + i0. e (}' . aµn (X) pa a 2 pa ., a pp - ~ 0 0Sa0Xa(x) + i0°a:a (apXa(x) + 8p(~e)paapwµ(x) + SpaPnP<\x) + e~eaP~=(x) - ~ e~eP a~Paµapnpa(x)) = eap mA (x) - e . npa(x) + 0(2'1' - + + S( 4ro - (x) - icrµa cp(x)) .,a ap ~~µ(x) i~aµx(x)) clA. µ µ + 2i88oµ wµ(x)- 2rneaµvµ(x) + i0°8P(crµppaµ(map(x)- mpa(x)) + ~ ~ 808(iaYCJµaµAy(X) +Ocp(x)) - ! 8 008(eap0mpa(x) - eapD n~a(x)) Separate map(x), na~(x) into symmetric and antisymmetric parts: map(x) = ~ eapm(x) + (O'vµe)apmµv(x), (47a) 18 pa _µv 1 ~J.1-,,V ~µ 1 pa . where m(x) = .!..e m ll(x), CJ =-(CJ cr -cr cr ) and mµv(x) = -(Ecrµ) m ll(x) is 2 at-' 4 2 a" anti-symmetric and self-dual, na~(x) = Ea~n(x) + eavµ E)a~ mµv(x) (47b) where n(x) =.!.e. na~(x) , cfv = .!..(cfcrv-av cf) and nµv(x) = 1 (EO'µ)h. n a~(x) is 2 ~a 4 2 "a antisymmetric and anti-self-dual. Then (38) implies that m(x) -n(x) = 0 (48a) 2'1'- c?~µ(x) + ic?aµx(x) = o (48b) 4ro - crµA.µ(x) - icrµaµcp(x) = 0 (48c) µ d Wµ(x)=O (48d) µ a Yµ(x)=O (48e) aµm(x) + aµn(x) = 0 (48f) iayc?aµ~v(x) - ox(x) = 0 (48g) icrVOµaµ"-v(x) + O = (crµ e):vµ(x) + (cfe):wµ(x) + 0p(2eap(l)a. - (cfe):A.µa(x) + i(cfe):aµ + e~(e~a'l'a + (cfe)~~(x) + i(cfe)!aµxa(x)) - i88crµa~aµmpa(x) - i88cr~~aµn~a(x) 1 · p-8( p _J.L...,.,V a µ _J)....,V a d V µ_p a v _p_µ a ) + 2180" (ea CJ O" E) a+ ea CJ O" E) a) µVv(x) - (ea CJ" O" E) a+ ea CJ O" E) a)dµ wy{x) + .!..8 8S. (Ea~D 19 ioµCf>a(x) -A.µa(x) + CJµaaroa. = 0 (49b) . a. . ioµXa(x) + ~µ(x)- ~ a:a'l'a = 0 (49c) _µ&p CJ aµm13a(x) = 0 (49d) cf. a npa.(x) = 0 (49e) a~ µ (Op ~v +cl clav)aµvv(x)- (avcfcl + avclcrµ)oµwv(x) = 0 (49f) . ~ µ a~ r/1. Dcpa(x) - i((j CJVE) aµAva(x) = 0 (49g) ea~ox'\x) - i(CJvcrµe)~0aµ~(x) = o (49h) (48c) and (49b) that (52) and (49d), (47a) and (48j) give Using -""=p) P = .l ( op) P ops:P (u CJ CJ 0 a 2 a +TI a tr(cfv) =0 tr(CJop cfv) = - .l(TlaµTlpv -TlavTIPµ + ieopµv) 2 and the self-dual, anti-symmetric nature of mµv(x), 0 apm p(x) = 0 (53) Similarly (49e). (47b) and (48j) give 20 µ a nµv(x) = 0 (54) Define (55) then (53) and (54) imply that µ () Aµv(x) = 0 (56) and also that ~ ecrpµv()pAµv(x) = ()Pm0 P(x) - ()Pn°P(x) =0 The latter implies that Aµv(x) is the curl of a vector, Aµ(x) say, Aµv(x) = oµAv(x)- ()~µ(x) (56) This exhausts the content of the massless superfield equation (38). Summarizing, we have found that the only propagating fields in the massless square root of the Dirac equation are a Dirac spinor cpa(x)) 'lf(X) = . ( Xa(x) satisfying the Dirac equation iPw(X) = Q and a complex vector Aµ(x) appearing gauge invariantly through · • Aµv(x) = oµAv (x) - ()vA(x) and satisfying the equation of motion µ a Aµv(x) = 0 Note that if the restriction to a Majorana-spinor superfield is made, then the field content is the same as that of the vector multiplet. Indeed the formulation (13) of the supersymmetric gauge theory gives a solution of (38); not of the constraints (23), however. A treatment which made more use of the superfield formulation, rather than the component field one given above, might reveal if, in fact, the two models are equivalent. There are a number of other aspects of the square root of the Dirac equation which might be explored. A treatment of the bosonic superfield would enable the massive case to be examined. Interaction terms which might be added to the free action 4 2 2 fd x d 9 d S ((B(x,9,S),A'P(x,9,S)) +Aa(x,9,S)M(~ .;) 21 References [1] J.Szwed, preprint, CPT, Marseilles (1986). [2] J.Wess and J.Bagger, "Supersymmetry and Supergravity," Princeton University Press (1983). [3] P.West, "Introduction to Supersymmetry and Supergravity," World Scientific (1986). [4] S.J.Gates, M.T.Grisaru, M.Rocek and W.Siegel, "Superspace," Benjamin/Cummings (1983). [5] W.Siegel, Phys.Lett. 85B (1979), 333. 22 Chapter 3 Supersymmetric Quantum Mechanics and the Index Theorem The Atiyah-Singer index theorem [1] is a result in differential geometry which was established in the sixties. Although the general result and its standard proof, which involves K-theory, are beyond most theoretical physicists, the statement of the theorem for specific cases is certainly not. One such case is the twisted Dirac operator or, in the language of theoretical physics, the Dirac operator in the presence of gravitational and gauge fields. In the seventies and eighties this case, together with some of its generalizations, was found to be of use in theoretical physics: primarily in the study of anomalies [2], but also in the analysis of the fermion spectrum in Kaluza Klein theories [3]. At the same time an index for supersymmetric quantum theories was defined [4] - the Witten index. It was a short step to see the index of the Dirac operator as the Witten index of an appropriate supersymmetric system, and then to attempt to arrive at the index theorem result by calculating the Witten index using the methods of theoretical physics [5]-[8]. Of course the proof that results does not have the rigour of the mathematicians' proof but it is comprehensible to theoretical physicists. In the first part of this chapter, we will review the Atiyah-Singer index theorem and give its statement in a number of cases - in particular, in the case of the twisted Dirac operator. We will then go on to show how this leads easily to the gravitational index therems for fields of arbitrary spin [9] and calculate the general expression in four dimensions. Finally, we will briefly review the applications of the index theorem to anomales and to Kaluza-Klein theories. In the second part of this chapter we will carefully derive the statement of the index theorem for the Dirac and twisted Dirac cases via the Witten index and using elementary path integral methods. While this has been done before [5]-[7], we will consider certain aspects in the construction of the supersymmetric quantum mechanics and in the construction and evaluation of the path integral at greater length. In particular, this will ensure that the normalization of the final result is properly determined through the calculation and not imposed at the end, and also that any other ambiguities which arise are properly dealt with. In the process we hope to have thrown some light on the path integrals themselves. 23 I. The Atiyah-Singer Index Theorem 1. Introduction The basic statement of the Atiyah-Singer index theorem is [l],[10] : Let M be a compact, smooth, n-dimensional manifold; let D be an elliptic operator on 00 00 M, D:C (E)~C (F), where E and Fare complex vector bundles over M; and let cr(D) be the symbol bundle of D. Then index D = (-l)nch(cr(D))td(TM®C)[TM] For the purposes of explanation it is easier to write this as . d D ( l)-in 24 k1 k,, Dk= ( 0:1) ··· (a~n) The index theorem requires that the operator D be elliptic, this means that the principal symbol of D, oi~> =I, ~ ~k. lkl=m where ~e Rn (or TP *(M)) and ~k = ~ 1 k!•• ~nkn, must be an isomorphism between the fibres of E and F for all pe M, ~:;eO. Obviously then, E and F must have the same dimension. We are now in a position to give the definition of the index of D. It is index D = dim(ker D) - dim(coker D). For an elliptic operator the kernel and the cokernel must be finite dimensional and so the index is well-defined. The index is also then given by dim(ker D) - dim(coker D*), where D* is the dual of D with respect to some inner products on COO (E) and COO (F). The right hand side of the index formula (2) consists of characteristic classes of vector bundles, with the whole evaluated on the base manifold M. Characteristic classes are objects in algebraic topology, in K-theory or cohomology. K-theory is the more suitable algebraic topology for consideration of the theorem itself, however, in the particular examples of interest it is through cohomolgy that the physical content becomes plain. A characteristic class is defined either for complex or for real vector bundles; of the characteristic classes occurring inthe index formula (2) the Chern character, eh, and the Todd class, td, are complex and the Euler class, e, is real. In the examples below these will reduce to other classes which will then be considered more explicitly through cohomology in terms of invariant polynomials in the curvartures of the vect~r bundles involved. The evaluation on M then just becomes an integral over M of the n-form part. The concepts of connection and curvature for vector bundles are more general than the like in theoretical physics. However, for the bundles derived from a tangent bundle with Riemannian metric, the connection and curvature can be brought into coincidence with those of gravitation, and for a vector bundle with a Yang-Mills type group with the Yang-Mills ones. That is the connection and curvature can be thought of locally as one- and two-forms respectively on M taking values in the Lie algebra of G and with certain transformation properties giving their global structure. Although the characteristic classes will be expressed in terms of the curvature, it is fundamental that they are concerned only with the vector bundle involved, and that they are quite independent not only of the choice of gauge or co-ordinate system but even of the connection used to define the curvature involved. That is they will change only by a total derivative, which when integrated over M, which is compact and without boundary, will vanish. 25 Summing up then, the left hand side of the index theorem formula involves the solutions to differential equations over M, while the right hand side involves only global topological quantities, and in fact, as can be seen from (2), the operator D only enters the expression through E and F, that is through the spaces between which it acts. There are a number of generalizations of the Atiyah-Singer index theorem [12]. For compact manifolds with boundary and non-compact manifolds the Atiyah Patodi-Singer index theorem gives the necessary corrections to the ordinary index theorem. Then there is the G-index theorem or character-valued index theorem. When there is a group G, with an action on M and a linear action on E and F which commutes with D, the g-index of D forge G is indexgD = trker D(g) - trcoer, k D(g). The G-index theorem gives a topological expression for this index. Finally, there is the family's index theorem, giving a topological expression for the index of a family of differential operators, a set of differential operators which are themselves parametrized by some compact manifold. 2. The Index Theorem .fur~ Exterior Derivative Qf Differential Forms Let M be an oriented manifold with even dimension, n=21. The bundle of differential forms over M, A*(M) = ESAP(M), p=O consists of the completely anti-symmetric covariant tensors at points of M. The exterior derivative operator d acts on the sections of A*(M), which we also call forms for convenience. Denoting by 1 (a,~)p = JMaA*~ . The dual of~ with respect to (, )P, (, )p+l is denoted op+1 and is given by op+l: C00(Ap+t (M))-7C00(AP (M)) 0 = -*d*. The operator d+8 on C00(A *(M)) is elliptic and complexifying A *(M), denoting its complexification A* (M)®C by . n A*= Ef>AP p=O 26 and extending d, B and* to A* we may consider the index theorem applied to the operator d+B: C00 (A"')~C00 (A"'). However, this is uninteresting as d+B is self-dual and so ind(d+B) = dim(ker(d+B)) - dim(ker(d+B)"') is trivially zero, as is the other side of the index theorem formula. . In order to achieve an interesting result we must split A* into two bundles of equal dimension, A*= EffiF, in such a way that d+B is off-diagonal 00 00 '(d+B)E: C (E)~C (F) ' (d+B); = (d+B)F Then ind((d+B)E) = dim(ker((d+B)E)) - dim(ker(d+B)F)). One way in which this may be done is by splitting A* into forms of odd and even rank. 0 dd I 2·1 1-1 2· 1 A* = A ffiA even = ffiA ffi ffiA i- i=O i=O Then the index theorem result 0 . d((d ~) ) =J ch(Aeven_A ~td(TM®C) m +u even M e(TM) can be reduced to ind((d+B)even) =IM e(TM) (3) where once again e is the Euler class. In terms of the curvature two-form on M, Rap• taking values in SO(n) 1 a1 .. · 00 ker BP= C (AP (M))\im dp-l and im dp-l clcer ~ so that ker(~+ BP) = ker ~ n ker BP = ker ~\im dp-l. and ind((d+8)even) = dim(ker(d+8)even) - dim(ker(d+8) dd) 0 = L(-1{dim(ker(d+8)p) p = L(-1 )P dim(ker cy'im d ) , p p-1 which is the index of the elliptic complex 27 0 00 p-1 <\.-1 00 p a., 00 0 ~ C00(A (M)) ~ ... ~ C (A (M)) ~ C (A (M)) ~ ... ~ C (A n(M)) ~ 0 It is only a slight generalization to write the index theorem in terms of elliptic complexes. Now ker dp/im ~-l is the pth de Rham cohomology group of M, HoRP(m;R), and is by de Rham's theorem isomorphic to the singular cohomology group of M, HP(M;R). L(-l)pdim ff(M;R) p is the Euler characteristic of M, X(M). Thus, in this case, the index theorem is just the Gauss-Bonnet theorem X(M) = JMe(TM) Another splitting of A* is into self-dual and anti-self-dual forms. That is setting .P(p-1)+1 Olp=l *p so that ro2 = 1 and ro(d+o) = -(d+o) , split A* into the + 1 and -1 eigenspaces of ro, A*=A+<:PJA- The index theorem result, in this case, ind ((d+6),) = Lch(A +_~~®CJ reduces to ind((d+o)+) = fML(TM) ' (4) where L is the Hirzebruch L-polynomial. It is given in terms of the curvature as -2 27tiR ) det1(tanhi!S._ , 21t which is to be interpreted as a power series in iR/27t. 1 2 21t 1 det ~ iRiR ) = 1 - --tr(RAR) + ... tanh- 241t 2 27t containing only forms of degree divisible by four. Using the fact that roP: AP ~ An-p, it can be shown that the contribution to ind((d+o)+) of forms of rank other than 1 cancel, and so ind((d+8)) = dim(ker((d+8\+)) - dim(ker(d+8)1) . Furthermore, if 1 is odd then ro1 is pure imaginary and the index vanishes. If on the other hand I is even, then ro1 = *1 and we need consider only real forms once again. The index is just the signature of ( , * ) on ker(d+o)1, or the topological signature of 28 M, 't(M). Thus the index theorem reduces to the Hirzebruch signature theorem 't(M) = JML(M). 3. The Index Theorem for the Dirac Operator Our last example is of most relevance to physics and in a sense the most fundamental. Let M be a Riemannian manifold which admits a spin structure and let A(M) be the spinor bundle over M associated with the Riemannian metric on TM. Further, let V be another vector bundle over M with group G. Then A®V, the twisted spinor bundle, is a vector bundle over M whose sections are spinor fields on M carrying a representation of G. Given a connection on V and the spin connection associated with the Riemannian metric on 1M, the Dirac operator is the map 00 00 iVy: C (A®V) -7 C (A®V) given locally in terms of an orthonormal basis {Ea} of TM , g(EwE~) =Bex~, as il\?' v'I' = EiVEa'I', for 'J'E C00(A®V), where V Ea is the covariant derivative in the Ea. direction and the product is Clifford multiplication. That is, i\?'v'I' ='fE~iVµ'I' where Eaµ are the components of Ea in a co-ordinate system {xµ}, Vµ is the covariant derivative in the ataxµ direction, -f are the Dirac matrices, {-f ,f} = 28cx~ , and 'I' is a Dirac spinor in an appropriate representation. It can be seen that iVv is elliptic, as its leading symbol ap(x,~)'I' = 'fE~i~µ'I' has square ~(x,~)'I' = -g(~.~)'I' = -gv~µ~v'I', which is just multiplication by a non-zei:o scalar for ~:;eO, since g is Riemannian (positive definite). The Dirac operator is self-dual on the full space C00(A®V) and thus its index trivially zero, so once again the SPJlCe must be split. If the dimension of M is even, n=21 say, then there is an element of the Clifford algebra the chirality operator r =l"(.l 1 ..."}l , such that 2 r = 1 and {,,a,r} = o, and such that splitting A(M) into eigenspaces of r, A(M) = A+(M)EIM.-(M), 29 i V'v is off diagonal, with (ilf'V+)* = i"1'V- • The index theorem result here reduces to ind~i~v+) = JMA(M)ch(V) (5) where A is the Dirac or A-roof genus A(M)=det;( ~ ) sinh! = 1 + l tr(RAR) + ... 2 12(47t) and eh is the Chem character again, given in terms of the curvature Q of V, which is a two-form on M taking values in the Lie algebra of G, as ch(V) = tr(exp(~)). This example of the Dirac operator on the twisted spinor bundle is fundamental in as much as the choice of the vector bundle V enables us to consider different spaces between which the operator is acting. In particular, by taking V to be. some bundle derived from the tangent bundle and with the connection associated with the Riemannian metric we may find the index of an operator acting on a field carrying other than the spinor representation of SO(n) - i.e. of spin other than a half. Remembering that by the ~ndex theorem the index depends only upon the spaces and not on the operator, the index found will be the same as that of the appropriate physical operator. V can also be chosen to be a vector bundle with Ga Yang-Mills group and Q the Yang-Mills curvature, or a tensor product of such a bundle with one as above, giving the index for an operator acting on spinors, or on higher-spin fields, carrying a representation of the Yang-Mills gauge group. 4. Gravitational Index Theorems for Arbitrruy Spin Fields Christensen and Duff [13] have considered the problem of index theorems for fields of arbitrary spin in four dimensions. Their method was to construct an appropriate physical operator and then to calculate its index by the heat kernel method. They met with difficulties in the shape of consistency conditions which imposed restrictions upon the underlying manifold and which ultimately, for higher spins, meant that the topological invariants and thus the index had to vanish. We give below a simpler derivation [9] based on the twisted Dirac operator and which encounters no such restrictions. 30 The Riemannian metric can be used to restrict the group of TM from Gl(n,R) to SO(n). This is usually done in terms of the frame bundle, whose points are th~ bases or the tangent spaces at the points of M - it is an example of a principal fibre bundle, one whose fibre is homeomorphic to its group. The restriction to orthonormal bases with respect to the Riemannian metric then reduces the fibre and group to S. The resultant principal bundle is called the orthonormal frame bundle - we denote it O(M). For every representation,cp , of SO(n) a vector bundle can be associated with th~ orthonormal frame bundle, just the vector bundle whose fibre is the vector space carrying the representation and which is twisted in the same manner as O(M) but through the given representation. Let us take V to be the vector bundle associated with O(M) by the representation cp. Let Iap be the antihermitian generators of cp, (6) [Jap'Jyo] = llaoJP'Y + llp/ao - lla/po -1lp0I ay • Then is the curvature of V, where is the curvature of M. The index of iV +is 1 (sinh iR J( ( )) I[q>] = tdet2 t ~" J exp 4~Ra\p (7) 41t To evaluate this explicitly we must take the terms of order 1 (n=21) in R, evaluating the trace of products of up to 1 Iap's. ·specializing to n=4, we have S0(4) = S0(3)®S0(3), and so the reporesentations of S0(4) are just the direct product of two representations of S0(3). The finite-dimensional irreducible representations of S0(3),,.SU(2) are of course just labelled by the non-negative half-integers. The representation labelled by s has dimension 2s+ 1. If La, a=l,2,3 , are anti-hermitian generators of s, ~, [La,Lb] = eabcLc • then tr(LaLb) = -ts(s+1)(2s+l)Bab. This can be seen by explicit construction of the La, or just by noting that in the adjoint representation tr(LaLb) = -28ab and for arbitrary s the coefficient is just tn" 31 A finite-dimensional, irreducible representation, cp, of S0(4) with antihermitian generators Ja~ satisfying (6) can be decomposed into a product of two finite-dimensional, i~educible representations of S0(3) by M ;_J J lT a-.!..Cb4 a e bc+-J42 a N a =~ 4 abe J be -lJ2 a4 ' then [Ma,Mb] = eabcMe ' [Na,Nb] = eabeNe . If the Ma and the Na generate the 2s1+1 and 2s2+ 1 dimensional representations of S0(3) respectively. Then cp denoted (s1,s2) is of dimension (2s1+ 1)(2s2+ 1), tr(JabJed) = eabeeedpCCMe+Ne)(Mr+Nr)) = eabeecdf(tr2s +1 CMeMf)tr2s +1 (l) + tr2s +1 (l)tr2s +1 CNeNf)) 1 2 1 2 = t<<>ad()be - <>ae<>bd)(2s1+1)(2s2+1)(s1(s1+1) + s2(s2+1)), tr(JabJc4) = eabdtr((Md + Nd)(Me - Ne)) = --j-eabe(2s1+1)(2s2+l)(s1(s1+1) - s2(s2+1)) and trCJaiM) = tr((Ma - NJ(~- Nb)) =-t-Oab(2s1+1)(2s2+1)(s1(s1+1) + s2(s2+1)), i.e tr(JapJ ) =i 32 1 I[s ,s ] = (1 + tr(RAR.)l (tr(l) + \ ( i ap tr(J AJ J 2 4 J~ AR~ i»J I 2 M 12(4n) JC 2. n) = ~2s +1)(2s +l)J ( 1 tr(RAR)- ~1 r(RAR)(s (s +1) + siCs +1)) 2 2 1 1 2 (4n) 1 M 12 3 1 ap 'YS ) +~apysR AR (s1(s1+1)- s2(s2+1)) =i<2s1+1)(2s2+1)( ! + s1(s1+1) + s2(s2+1))P(M) (9) +-k<2s1+1)(2s2+1)(s1(s1+1) - s2(s2+1))X(M) where P(M) = 2 f Rap AR (41t)2 M Pa is the Pontrjagin number of M, and l f ap yS X(M) = i E p 0R AR 2(41t) M a "f is the Euler number of M. This then is the general result for the index of an elliptic operator between fields differing by integral spin. The first few cases are: I[0,0] = f4 the index of the ordinary Dirac operator, related to the spin-1/2 axial anomaly (see below); the index of an operator C 00 (~(l ,O) 33 5. The Index Theorem filld Physics The primary application of the Atiyah-Singer index theorem to physics is in the study of anomalies [2]. These occur in the transition from classical to quantum theories; a symmetry of a classical theory may not be preserved by quantization, leading to changes in the associated identities. In the most extreme cases anomalies can even prevent the construction of a consistent quantum theory, as for the non abelian anomaly in chiral gauge theories and the conformal anomaly in string theories. Anomalies can often be related to one form or other of the index theorem, and this, besides providing a method for their calculation, shows their essential topological nature. Another important application of the index theorem and of its generalization the G-index theorem is to the analysis of the spectrum of massless particles in Kaluza Klein theories [3]. The demonstration that the apparent chiral asymmetry of nature could not be produced in a pure Kaluza-Klein theory contributed largely to the demise of such theories. The original and simplest anomaly is the axial U(l)-anomaly [15]. By using Fujikawa's method [16] the relationship to the index theorem can be easily seen. Let 'lf(X) be a Dirac fermion in a 21-dimensional Riemannian space time with metric gµv(x), and in the presence of an abelian gauge field Aµ(x). Then the classical action is. 21 S(A,g,'lf,'lf) = Jdx'lf(x)iV'lf(X) , dx = d xfg It is invariant under a global chiral transformation, 'lf(X) ~ eicxr'lf(x) 'lf(x) ~ 'lf(x)eicxr ' since {r,V} = 0. Under· a local transformation, a.= a.(x), S ~ S + Jdxa.(x)Vµ.i~(x) and thus the axial current, j~(x) =wfnl' is conserved classically - v~~(x) =0. For the quantum theory, Fuji~awa's method is to examine the path integral · - Z(A,g) = J[d'lf][d'lf]exp(-S(A,g,'lf,'lf)) ~his integral should be invariant under a transformation of the integration variables 'lf,'lf, provided the change in the path integral measure is taken into account as well as the change in the integrand, 8 i.e. Ba.(x) Z(A,g) = 0 (10) When the space-time is compact, iV' has a dis~rete spectrum of eigenvalues and 'If ,'If can be expanded in terms of a complete set of orthonormal eigenvalues: 34 'lf(X) = Llln'l'n(x) , 'lf(X) = L'l'~(x)bn n n where iV'lfn(x) = An'lfn(x) , Jdx'lf!(x)'lfm(x) = Bnm . Then the measure can be defined as [d'lf][d\jf] = fldbndan. n Under an infinitesimal local chiral transformation 'lf(X) -+ 'lf(X) + icx(x)I\jr(x) 'lf(X) -+ 'lf(X) + iCX(X)'lf(X)f' , ~-+ ~ + Lamfdx icx(x)'lf~(x)'lfm(x) m bn-+ bn + LbJdx icx(X)'lf~(x)'lfn(x) m and the Jacobian factor is (a~'I (ab 'I def\a~) def\ab:) = exp(-2ifdx cx(x)~'lf~(x)f''l'n(x)) , where a suitable regularization is assumed. Now, if then so that, for An:;i!:Q, On the other hand, on ker(iV) [i?',f'] = 0' so that the zero modes of i?' can be split into those of positive and t~ose of negative chirality, and thus : fL'l'!(x)I\jrn(x) =ind(iV +), n the index of the Dirac operator on the space of positive chirality spinors. The index theorem (5) can now be used to give the index as a space-time integral, and then for nearly constant ex Ja(x)L'l'~(x)I"lfn(x) = Jdx cx(x)A(x)((F) n ~ ) = fdx (4+N)-dimensional manifold M4xB, where M4 is conventional space-time and Bis compact and of the scale of the Planck length, and on which there are defined a metric tensor and matter fields but no elementary Yang-Mills gauge fields. Continuous symmetries of B will manifest themselves in four dimensions as gauge symmetries with the corresponding gauge fields emerging from the metric tensor. The observed matter fields in four dimensions must originate from massless fields on M4xB as all other fields have masses of the order of the Planck mass. Further, for Dirac fermions the Dirac operator on B will act as a mass operator in the four-dimensional Dirac equation. Thus the spectrum of observed particles must emerge from the zero modes of the Dirac operator on B. Now the observed spectrum is chirally asymmetric both in the relative number of pru:ticles of each chirality and inthe representations of the gauge groups which they carry. Thus the 0-index of the Dirac operator on B must be non vanishing. Unfortunately by a theorem of Atiyah and Hirzebruch [19] this is never so. As a result the only physically interesting theories are those such as the superstring theory which begin with elementary gauge fields and matter fields in chirally asymmetric representations. 36 II. Derivation of the Index Theorem via f!: Supersymmetric Quantum Mechanics 1. Preamble There are a number of derivations of the Atiyah-Singer index theorem for the Dirac operator which use the methods of physics. One obvious approach is to turn the relation to the chiral anomaly around and then to calculate the chiral anomaly by some other means. This requires calculations in a quantum field theory on the manifold involved. Simpler methods are based on an analogy between the index of a differential operator between the sections of two vector bundles and the index of a supersymmetry operator between the bosonic and fermionic Hilbert spaces of a supersymmetric quantum mechanics - the N=l/2 supersymmetric non-linear sigma model being the model appropriate for the Dirac operator. The latter index can be written as a trace of the evolution operator and then evaluated. The advantage here is that the calculations are only of a quantum mechanical model, that is a one-dimensim~al quantum field theory. The usual method of evaluation is via a path integral formulation of the trace [5]-[7], although Zumino [8], referring to the trickiness of the definition of the path integral, opted instead for a WKB calculation. There are also rigorous mathematical evaluations of the trace [22] including one which uses Wiener integrals, the mathematical equivalent of path integrals. As stated at the beginning of this chapter, we will consider the path integral approach. First, though, we review Witten's supersymmetry index itself. 2. Witten's Supersymmett:y Index The concept of. the index of an operator was introduced into the study of supersymmetric theories by Witten [4]. He argued that the index of a generator of a supersymmetry must. be zero if that supersymmetry is to be broken. Since the spectrum of observed particles does not consist of degenerate boson-fermion pairs, any physically realistic supersymmetric theory must have its supersymmetry broken. This requirement is equivalent to saying that the energy of the vacuum must be non zero. However, this may be very difficult to check precisely when the symmetry breaking arises from quantum corrections. The requirement that the index be zero, although not a sufficient condition, has the advantage that the index, being a topological quantity, can be calculated exactly in some convenient limit. To show how the condition comes about it is sufficient to consider a one dimensional or quantum Il!echanical model with one supersymmetry generator S acting on the Hilbert space HBEBHp, with 0 Qt) s = ( Q 0 ' S2=H, the Hamiltonian, and 37 (-ll = (6-~), where F is the fermion number operator. The states of interest, the zero-energy states, arise for a four-dimensional model within the.zero- momentum subspace, on which the supersymmetry algebra reduces to A- A {~,~B} = 2BB5a~H, justifying the use of the simpler model. A symmetry is unbroken if and only if its generator annihilates the vacuum, s 10) = 0. In this case, that implies that the vacuum has zero energy H 10) = 0. On the other hand, if the supersymmetry is broken, then s 10) :;C 0 and the positivity of H = s2 = sts implies that the vacuum has energy greater than zero. Thus, if there is a state with energy zero, then the vacuum being a state of lowest energy also has energy zero. From the above then, the supersymmetry is broken if and only if ker H = ker S = {0} and, since t ker S = ker Q Ee ker Q , it is a necessary condition for broken supersymmetry that the index of Q, ,. ind Q = dim(ke~ Q) - dim(ker Qt) be zero. Or, in other words, if the index of Q is not zero, then the supersymmetry must be unbroken and the model is not a physically realistic one. The index of Q can also be written as the trace over the zero energy states of (-l)F, i:o<-ll. A more convenient form would be as a trace over all states, this would be an infinite sum so that care would have to be taken to ensure that it converged. To realize it note that for each non-zero mode hjf), with Hl\jl) = El\jl) :;e 0 , there is another state Sl\jl) :;e 0 with the same energy, HS l\jl) = SHl\jl) = ES l\jl) , and opposite statistics, F F (-1) Sl\jl) = -S(-1) l\jl), so that (12) 38 Provided that the spectrum of His suitably distributed this infinite sum will be well defined and independent of (3>0. It provides the basis for the calculations of the index. 3. Formulation of the Path Integral For a quantum mechanical system with euclidean time and hamiltonian H, the operator exp(-(3H) gives the evolution of that system over time (3. The path integral representation for matrix elements of the evolution operator in such a system is well known for conventional time [20], whether the quantum mechanics is given in terms of position and momentum or in terms of bosonic or fermionic creation and annihilation operators. Below we briefly review the euclidean time formulation in its simplest case - a fermionic system with one degree of freedom - arriving at the expression for tr((-l)Fexp(-(3H)) as well as for tr(exp(-(3H)); we then state the result for the other cases. A fermionic system with one degree of freedom has a two-dimensional space of states. _At a given point in time it has an orthonormal basis {10), 11)} given in terms of the creation and annihilation operators, a and a, which satisfy -a = a t , {-a,a } = 1 , _2a = 0 = a2 , by alO) = 0, alO) = 11), (OIO) = 1. An arbitrary state If) can be given as either If)= alO) + '311) or If)= (a+ (3a)IO) and an operator A as The analytic functions of a single complex Grassmannian variable, T\. provide a realization of this system. The state If) is given as f(il) = a + 1311 and the actions of a and a are realized as af(r\) = T\fCf\) and afCf\) = d~fCf\) . The scalar product is (f lf ) = Jdooa e·a31 (a)f (a) . 1 2 1 2 To an operator A can be associated two functions ofT\,T\: the kernel ACf\,T\) = Anmllllrim; n,mL and the normal kernel They are related by 39 Tiri N A(i'\,T\) = e A (ii ,T\) . The action of A is given by (Af)(il) = fd~dT\ e {riA(il,T\)f(~) , the trace of A by tr(A) = fdT\dtl eilTIA(il,T\) (13) and the product of two operators A1 and A2 by - -!;!; - (A 1 ~)(il,T\) = Jd~d~ e A 1 (i'\,~)A2 (~,T\). (14) Let the hamiltonian be given in normal ordered form, so that we have its normal kernel h(T\,T\) =HN(T\,T\). To find tr(~xp(-PH)) we require the kernel of exp(-pH). Now splitting the time interval p into N intervals .6.t = PIN. with N very large, we have e -PH = (e -AtH)N and e-AtH = 1 - .6.tH . So th~ norm~ ~ernel of exp(-.6.tH) is approximately 1 - .6.t h(1l,tj) = e-At h(Tj,ri) and the kernel fiTI -At h(Tj,T)) e . Then using (13) and (14) tr(e-PH) = fdT\dtl eTiTl(e-PH)(i'\,T\) dT\dtlftdrt.dT\. exp(1lT\ - !rt·T\· + TlT\N-l - .6.th(Tt,T\N-l) = f i=l 1 1 i=l 1 1 + !01.+11'\· - .6.th(1l. l'T\.)) + 1'1111- .6.th011'11)) i=l 1 1 1+ 1 = JdT\drtf1drtid11i exp(Tl(11+11N_1) -.6.t h(i1,11N_1) . 1=1 -6 So that taking 11 = 11N• T\ = 11o = -11N and the limit N-700, .6.t-70, we have the path integral representation tr(e-PH) = JIT dil(t)d'll(t) exp(-fdt (1'11'\ + h(i''i.11))) (16) APBC t O where APBC stands for anti-periodic boundary conditions. It is to this limit that we will return later to explain the ambiguities arising in the path integral calculations. We also require tr((-l)Fexp(-~H)). Now (-l)F = 1 - 2aa so that its normal kernel is - -2TjT)Nl 1 - 2111'\N-l = e 40 and its kernel e-fiTIN1 . Thus tr((-l)Fe-J3H) = JdT\dfifidfjidT\i exp01(T\-T\N-1) - !01i+l(T\i+l-T\i) + ~t h01i+1 'T\i)) i=l 1=1 -fi1<1'l1-T\)- ~th(fipT\)) so that taking ii= -TlN• T\ = T\N = T\o• we have tr((-ll)e-J3H) = J IT dfj(t)dT\(t) exp(-fdt 011'\ + h(fi,T\))) (17) PBC t 0 where PBC stands for periodic boundary conditions. All of the above of course generalizes straightforwardly to a fermionic system with multiple degrees of freedom. For the actual evaluation of such path integrals a different realization of the path integral measure is used - we will use one in terms of modes of Fourier expansions of T\(t), T\(t) on the interval [O,p]. For this reason it is important to fix the normalization of these path integrals. For a fermionic system with d degrees of freedom (16) is easily normalized using i.e. Note also that tr(e-J3H) = tr((-ll(-ll e-!3H) = IJdfi(t)dT\(t) exp(-2fi T\ )exp(-fdt 011'\ + h(fi,T\))). Jmet 0 0 o Splitting the integral into a conventional integral over constant modes, fdT\dii, and a path integral over non-constant modes which vanish at the boundaries, denoted f0DliDrt. we have tr(e-J3H) = JdfidT\ exp(-2fiT\)f DfiDT\ exp(-fodt 011'\ + h01,T\))) 0 and, when H = 0, On the other hand, tr((-lle-j3H) ~ fdfidT\f DfiDT\ exp(-fodt 011'\ + h01,T\))) . 0 Combining the two, F -j3H dfidT\f DfiD11 exp(-fdt .011'\ + h01,T\))) f 0 0 tr((-1) e ) = - (18) DfiDT\ exp(-fofii\) f0 41 The path integrals for bosonic systems are analagous - that is the exponent is in each case -SE, where SE is the euclidean action corresponding to the given hamiltonian. For a bosonic system defined interms of creation and annihilation operators tr(e-PH) = f rrdf\~~T\(t) exp(-fdt (T\1'\ + h(T\,T\))) . (19) PBC t 1tl O In terms of position and momentum tr(e-PH) =:= Jrrdp(~~(t) exp(-fodt (-ipx + h(x,p))) t where the path integral is over those x(t) with periodic boundary conditions. When the hamiltonian is of the form 2 h(p,x) = 1- 42 4. The Index of~ Dirac Operator Let M be an (n=2l)-dimensional riemannian manifold on which is defined a spin structure. The index of the Dirac operator over M can be expressed as a path integral in a supersymmetric quantum mechanical system through the identifications: C00(A(M)) H H r H (-l)F 00 C00(A +(M))(f)C (A-(M)) H HB(f)HF - 1-fi( H S .fi .!.. (i~)2 H H 2 That is, by finding a quantum mechanical space of states isomorphic to the space of spinors on M, the index of the Dirac operator can be seen as the index of the corresponding quantum mechanical operator which generates a supersymmetry. Its index can then be expressed through tr((-l)Fexp(-J3H)) as a path integral. The fact that this path integral must be independent of J3>0 then enables it to be evaluated. The appropriate space, H, is the direct product of a position-momentum space with n degrees of freedom and a fermionic space with 1 degrees of freedom. The position space represents the points of M and the ferrnionic space the spinor structure. So that 'JIE COO(A(M)) is given as 1 l'Jf) = Jdx jg 'I'· . (x)lx)@i •••fi . O) (22) J1 ···Jc 1 1 The representation of the if-dimensional space of spinors at each point in terms of I fermionic creation operators, aj, j=l,... ;f, comes about in the following way [10]. Given the complexified Clifford algebra, C21®C generated by "f, a=l, ... ,2t {"f,"f }=28a~, the space of spinors can be realized as a left ideal of C21 ®C, A = {'JIE C21 ®C; Qj'Jf=-'Jf, j=l,... ,.t} , where ~ is right multiplication by ifj-lfj , j=l,... ,l . The Clifford algebra action on the spinors is then just Clifford multiplication from the left. Now writing 2' 1 2' a. = .!.. ("( J- + iy J) J 2 -a.= - 1 ( y2j-1 - 1y. 2j) J 2 we have {aj,ak} = 8jk, {aj,ak} = o , {aj,ak} = o . Further the 21 products TI{a.,1-a.a.} , j=l J J J where for each j we take one of the terms in brackets, form a basis for the space of spinors. Thus defining 43 so that a.10) = 0 , j=l,... ,t, J the space of spinors is the space of states generated from 10) by thelcreation operators aj. In this representation y2j-1 =a.+ -a. J J 2j .( - ) y = -1 a. - a. 'j=l,... ,f, J J and .1 1 -~ r = 1 y .. :y = IT (1 - 2a.a.) . 1 J J J= = exp(-2~).a.) . J J F J = (-1) . The Dirac operator is i~ = 'fE~(x) (idµ+ iroµ(x)) where Eµ 0 '(x) is the inverse vielbein, ap µ v µv B Ea(x)EP(x) = g (x) and roµ(x) = lro A(x)[rfX,/] 8 µal-' is the spin connection. 2 ; (i~) = ! {'f,/}E~(x)E;(x)(idµ +iroµ(x))(idv+irov(x)) -! 'f/E~(x)E;(x) [V µ.V vl =; gµv(x) (idµ+ iroµ(x)) (idv + irov(x)) - 1 'f/E~(x)E;(x)y'YfRµvyo(x), 16 using [V WV vl = ~ Rµvyo[y'Y,f] , = ; gµv (x) (idµ+ ~roµ(x)) (idv + i(l\,(x)) + ~ R(x) , where R(x) = E~(x)E;(x)Rµv ap (x) is the Ricci scalar. In the quantum mechanical representation -idµ~ Pµ so that l(i~°)2 ~ H = lgµv(X) (Pµ-liro A(X)'f/)(Pv -liro ,,(X)y'Yf) +lR(X) . 2 2 4 µal-' 4 V'(u 8 In order to substitute this hamiltonian into our expressions for the path integral we must ensure that the terms involving aj, aj are in normal ordered form (on the other 44 hand, we will assume that the ordering of P and X is correct, see Faddeev [20]). Then defining oa~ by 2j-l,2j A 2j,2j-1 n = 1 = -u andVX by 2j-1 1 ( - ) "' =- 11. +11. . .fi J J "'---11.-11.'2J - i ( - ) .fi J J we have ind(i~ ~ = tr((-l)Fe-~H) = J IT!i dxµ(t) dlf Ct) d11j(t) PBC t exp [-J:dt (~ gµy(x):ilxv + ~ xµ coµa~(x) Cfi + ~ Qa~) + lfi\j + ~ R(x))] Making the change of variable _j . 11 ' 11J -7 'l'a _j . 2· 1 2· df\ (t) d11J(t) = -i d'I' J- (t) dv J(t) . . 2· 1 2· __.). J 1 ( 2j-1. J- 2j. J d ( 2j-1 2j)) 1111 = -2 "' "' + "' "' + -dt 'I' "' ' 1 ind(i~J = J IT!i dxµ(t) c-i) dvact) PBC t dt (gµv(x)xµx v + 8 Rnp + R(X)'l'p +i. xµ (I) R(x)Q R(x))] exp[.!..J~, 2 o a.., vCXXµ~µa.., 2 µa.., Cl~+_!_4 (23) Where the curved background introduces ..../g into the measure for xµ(t). The evaluation. of the integral proceeds through the separation of the integration variables xµ(t), vaCt) into constant and non-constant modes as in [5J,[6J xµ(t) = x~ + uµ(t) , uµ(O) = 0 = uµ(J3) v«Ct) =vg + ~a(t), ~a(O) = 0 = ~a(J3), with the normalization of the integral as given earlier. That is 1 1 ind(i\f+) = (-i) Jd~ (21tf3f Jdx~ Jg(x ) J o~aJ Duµ 0 0 0 exp[-.!..Jpdt (gµvxµxv + 8 Rn~+ 'l'CXXµ (I) R(X)'I'~ +i. xµ (I) Rna~ + .!..R(x))] 2 o a.., µa.., 2 µa.., 4 x ~OD~ aexp[-u: dt 0.. l~p] JoDu" expH J: dt 0apu "uP] r where Duµ= Ilduµ(t)J g(x +u(t)) . 0 t 1 1 2 -7 -7 , -7 Under the rescaling t J3t, 'l'o (3-\,0 u(t) J3 u(t) , 45 the measure becomes 1 ) ) (21ti) · J d'lf~fdx~ Jg(x 0 J D~ ex JfI duµ(t)J g(x0 0 t -1 (24) • • I x~ Dt;" ex{i f~dt oalf!] fDu" ex{i f~dt Oapu"uP + O(~~ 0 0 J) and the exponent becomes 1 Jld ( ( ).µ.v ~ i:cxi:P .µ.v ( '\ cx,..,P 2 o t gµv xo u u +ucxp~ ~ +u u roµ,vcxp XCY'l'o'l'o ' 1 2 + 2'1fgit (l)µcxp(Xc)'lfp(t)) + 0(p ) (25) Now recalling that the index must be independent of p, we may ignore the terms of order "1p. Note that then the terms . µ exp 1 .!... co +-R(x) , 2 x µa..,Rn 4 which arose from a correct consideration of operator ordering but were neglected by other authors [5],[6], do not contribute as we would expected. If we also make the substitution [5] ~cx(t) -7 ~cx(t) - ro:~(x0)"'°uµ(t), then the measure (24) is unchanged and (25) becomes dt (gµv(x )luv + + B -~ J~ 0 ~ Rµvcx~(x0)uµuv~"'° ap~cx~p) and, setting ucx(t) = e~(x0)uµ(t) , we arrive at the final path integral expression: -1 1 ind(il7.,) = (2ltif J d~Jdx~ g("o) ~Du" ex{ iJ~ dt ua :; u"] J JDua exp[.!.. J1 dt ua(£ B R - lR R.. (x )'1'~~.Q._luP] (26) 0 2 O dt2 ex.., 2 ex.., 11> 0 dt) That is the only path integral remaining to be evaluated is "gaussian". The ordinary gaussian integral is 1 fdx e·xtAx oc: det2A In (26) the matrix of which we must take the determinant is infinite dimensional and care must be taken. Expanding ucx(t) in a Fourier series 46 ua(t) = L(aa/2 cos(2mtt) + ba/2 sin(2mtt)) n=l and writing formally 0 x1 -xl Q_ ·. x 0 1 -x1 0 2 .!L B rt - l R rw"'I'~~o takes the nearly diagonal form dt2 a.., 2 a.., 1u 0 0 -27tn -x. 1 0 0 x. -27tn co 1 1 Ef) Ef) n=l i=l 27tn -x. 0 0 1 x. 27tn 0 0 1 and This infinite product is standard and its limit is 1 Changing ~ ~ ( 2~)°2'1'a, 47 (l (ll I 21 tinhiR ind(iV+) =JMdet ~1t J= JMA(M) , 41t which is the Atiyah-Singer index theorem result for the Dirac operator (5). 5. The Index of the Twisted Dirac Operator We now consider the index of the twisted Dirac operator, iVv+:C 00 (~ +(M)®V) ~ C00(K(M)®V), i.e. the index of the Dirac operator in the presence of gauge as well as gravitational I -- _. fields. Let G be the group of V, the gauge group, cp the representation involved and d the dimension of cp. The quantum mechanical space of states isomorphic to COO(~(M)®V) consists of states J1 J1 - l'!f) = dx./g ~ . (x)lx)®a ••• a IO)®calO) f J1 ...J1 1 1 = l'!fa)®calO) where 'l'a carries the representation cp, and each l'lfa), a=l, ... ,d, is an element of COO(~(M)) as given in the last section. The ea generate a fermionic (although bosonic would do just as well) space of states through {ca,cb} = 0, {ca,cb} = 0, {ca,cb} = 8~, calO). Note that in order to consider only spinors 'l'a carrying the representation cp, and not all of its anti-symmetric (symmetric) products as well, the space of states must be restricted to the "one-particle" set of states - those generated from the vacuum by just one Ca· The twisted Dirac operator is iVv = i'fE~(x) (aµ+ roµ(x) + Aµ(x)) where Aµ \(x) is the Yang-Mills connection taking values in the representation of the Lie algebra, g, of G corresponding to cp. From now on we will supress the gravitational terms as they do not mix with the Yang-Mills terms. 2 ; (iVy) = -; gw eaµ+ Aµ(x)) (av + Ay(x)) - ! Y''-lPv µ.V vl = ; 8µv(iaµ + iAµ(x)) (iav + ~(x)) - ! Y''lFµv(x) 48 where Fµv(x) is the Yang-Mills curvature corresponding to Aµ(x). In the quantum mechanical representation, an element of q>(g), T\ is replaced by eaT1 bcb so that carbcYcclO) =Tb'l'~alO). Representing xµ, -iaµ> 'f as before 2 2 ~ (iVy) ~ H = ~ (Pµ - iT'iaA; b(X)Tlb) - ~ ~'l'vFµy(X). The index is again given by a trace, ind(iY'v+) = tr 1 ((-l)Fe-~H), (28) where the subscript 1 reminds us of the restriction to the one particle states of the c fermions, and r ~ (-l)F, that is Fis here the number operator only for the a-fermions generating the spinor space. There is now a problem in constructing a path integral formulation for this trace, as the standard procedure requires that the trace be over a full set of states, not just the one-particle states. Some authors [5] have found it sufficient to merely attach an apostrophe to the relevant integral and proceed with the path integral manipulation untroubled. In a situation such as this, where the correct answer is already known anyway, such an ill-defined approach seems unsatisfactory. An alternative approach [ 6] is as follows. Extend the quantum mechanical space of states to the full set, but rather than taking tr((-lle-~H) take ,tr((-lte-~H +iaNc) where Ne= 'L,caca is the number operator for the c-fermions. Then on the one hand c tr((-lte·~H+iaNc) = !einatrn((-lte-~H) (29) n=O where trn is the trace over the n-particle states for the c-fermions, and, on the other hand, the full trace may be written as a path integral with modified hamiltonian, H' = H _ia N ~ c The situation is not quite that of the Witten index, however, the arguments for independence of~ still apply since {(-l)F,S} = 0 and [S,H'] = 0 still hold. After the path integral has been evaluated, it may then be expanded as a power series in exp(ia) and the index tr 1 ((-l)Fexp(-~H)) extracted. Note also that the n=O term in (29), tr0 ((-l)Fexp(-~H)) is just the index for the ordinary Dirac operator. Writing 11a,;;a for the Grassmann variables corresponding to ca,ca in the path integral, " 49 1 tr((-lte-~H') = J IJ dxµ(t)(-i) dv«Ct)J IT dfia(t)dT\a(t) PBC t APBC t 2 exp[-J:dt (; x + xµfiaA: b(X)T\b - ; o/\1,VfiaF µ~ b(X)T\b - ! QµvfiaF µ~ b(X)T\b (30) where anti-periodic boundary conditions arise for the T\,Tl integrations and nµv is as in the previous section. Expanding rescaling (31) ' where the gravitational terms have been reinstated. Taking care of the ~a. and uµ ) integrations as before and denoting ; 'V~'lf~Fµ~ b(x 0 by F ab , we have .. 1 jsinhRJ 1 2 2 = (21ti)- Jdx~.{g" Jd'lf~ det L R 2 xJAPBCI;l di],(t)dT)'( t) ex{; J~ dt 'i]~tt o: -F 'b -iaf b] (32) Once again the independence of f3 given by the supersymmetry allows the original path integral to be reduced to a product of gaussian path integrals. Formally diagonalize the anti-hermitian matrix Fa = (zl ·.. ) b zd and expand T\a(t), T\/t) in a complex, anti-periodic Fourier series so as to diagonalize d dt ' 50 a() V a (2n+l)im 11 t = L,.; 11n e n=-oo - ( ) V - -(2n+l)im 11a t = L,.; 11na e n=-oo The 11,11 integral then becomes Jij dffnad11~ exp[t ((2n+ l)i1t - Za - ia) ffna11~J =(I ((2n+ 1)!1t - Za - ia) 00 a=l Unfortunately, even after normalizing with respect to the free integral, this product is divergent. detAPBc(-£t- F ab - ia) II(1 + i Za+ia ) = (33) det -d J n,a (2n+ 1)1t APBC (dt 1 has the same convergence properties as L n+ . n=-oo 2 1 Using some regularization procedure we could re-arrange the terms of the product so that we had 2 . za+ia . za+ia (za+ia.) . ···. llllT1-in-( 1 +l x1 +I )-ITd- 0( 1+ J. a=l n=O (2n+l)1t (2(-n-l)+l)1t a=l n=O (2n+1)21t2 This product is well-defined and = d cosh(za+ia) -- 2 rra=l = det (cosh(~ (F +ia) )) . However, this does not give the correct form for the index, indeed it may not even be expanded in integer powers of exp(ia). Other re-arrangements of the terms of the product will yield other results. Let us examine the path integral itself in order to find the origins and the extent of this ambiguity. Let us consider a fermionic system with one degree of freedom and hamiltonian H = -zaa. The trace of exp(-H) is easily evaluated in this case. The space of states is spanned by 10) and a 10), and, using (aa)n = aa' exp(zaa) = 1 + (ez-l)aa and tr(exp(zaa)) = 2 + (ez-1) = 1 + ez. (34) On the other hand, section 11.2 enables us to write 1 tr(exp(zaa)) = J IT dff(t)d11(t) exl-J dt (Tfl'l - zfi11)] APBC t l'L 0 51 Attempting to evaluate this path integral without recourse to its origins leads as above to (35) which is ambiguous. We may, however, make the following arguments to restict its value [6]. We assume that it is an entire function of z, that is analytic over the whole complex plane, and denote it by f(z). Now, the product is only zero if one of its terms is zero, that is f(z) has zeros only at z = i(2n+ 1)7t. This immediately implies that f(z) = eg(z)cosh(~j (36) where g(z) is an entire function, determined by the behaviour of f(z) as lzl-700. From (35), f(z) ~ 1 f(z) = z-i(2n+l)7t ' n--oo~ assuming that such manipulations are meaningful, as well as by (36) being = g'(z) + ttanh(~j. Its seems reasonable to assume that this is bounded as lzl-700 away from z=i(2n+ 1)7t, and thus that g'(z) is bounded. Therefore g'(z) , being bounded and entire, is constant: g'(z)=b, say, and 'g(z)=a+bz, f(z) = Nebzcoshl~J (b.l.)z =Ne z (1 + ez) 2 Thus the path integral is determined only up to an overall normalization, and, assuming b to be real and z imaginary, a phase. The origins of this ambiguity are in the discrete to continuum limit used to derive the path integral. We may confirm explicitly that the discrete integral formulation of the trace is exact. That is (15) exp[ii<11+11N-l) + iiTlN-1 - !(11i+1<1li+l-11i) - NZ ili+l'lli) ~fd11diiftdilid11i1=1 ~ 1=1 \ - i11<1l1-1l) + ~ 11111] 1 = JJ.~~,J(-l)N- d11di1rf dnidili exp[il'Tl + i1TlN_101f) ·_ 11111 1 + i11n01f) + 6(11i+l1li(l1f) -ili+lili+1)] 52 1 14 N . 14 -1 N = lim (1 + (14-)N) N~ N = l+ez = tr(exp(zaa)) On the other hand, other discrete integrals seemingly also converge to IT df\(t)d11(t) exp[-J1 dt (T\ft - zf\11)] JAPBC t 0 such as ~fd11df\ndf\id11i exp[f\(11+11N-1) + 2~ f\(11+11N-1) - 6(fli+1<11i+l-11i) - 2~ fli+1<11i+1+11i) )-111<111-11)+ 2~ fl1<111+11)] l 12~ 1+4--2N N-1 = lim (-1) det 112~ -112~. N~ . z z =e2+e2 = coshCZ) 2 Thus it seems that information is lost in the change to the path integral and, taking these two examples as a guide, the ambiguity is as derived earlier. Returning to the derivation of the index, we have JAPBJ.l d'i'j,(t)dTj'(t) ex{ J. dt 11.(~1 s:- F \ - i~f b] (i (b-1)(F +ia) . ) = det~ e 2 (1 + eF +m) In order to determine what values N and b should take in this particular case, Windey [6] assumes that this path integral should take the same value as the corresponding trace, that is 53 However, if, as we have seen above, the ambiguity arises from the many to one nature of the change from the discrete to the path integral, then after manipulations have been performed upon the path integral we are not justified in returning through a discrete integral to a trace - that is unless we write det(~ exp((b- ~ )(F +ia;))}i-(exp(Ca(F ab +ia.8ab)cb) Singh and Steiner [21] give a simple example of how a manipulation of a path integral may lead to incorrect results. Using integration by parts ld -· Jld - J 0 t 1111 = - 0 t 1111 but, on the other hand, r.11i+1<11i+l-11i) "#-r.<1li+l-1li)1li+l. 1 1 They go on to evaluate the matrix elements of the time evolution operator for the quite simple hamiltonian which they are considering using only the discrete formulation. Unfortunately, such an approach is not practical in the case with which we are concerned due to the complexity of our hamiltonian. Any attempt [9] to evaluate the index through a discrete formulation must become hopelessly bogged down in the algebra. The reduction of the initial path integral into integrals over ... constant modes and gaussian integrals over non-constant modes does not have an obvious equivalent in the discrete formulation. In fact in this case we can fix the values that N and b should take in order that 1 sm. hRJ- · F ~H· 1r µ r -2 2 tr((-1) )e- ) = (27tif Jdx fg Jd-VXdet R ~ 2 det [~xp((b-; )(F +ia;))(l + exp(F +ia;))J (3?) be true, without appealing to any a posteriori argument. We saw at the beginning (29), that tr((-l)Fe-~H') also equal to Leinatrn((-lle-~H), n=O with the coefficient of eia being the index which we are seeking, and the constant (n=O) term being the index for the ordinary Dirac operator, which we have already found. Now (37) may only be expanded as a series in positive integral powers of eia if b=112.3!2 ... det [~xp((b- ~ )(F +ia))(! + exp(F +in))J = ( ~ Je(b :)dadet e (h :I' (1 + e"'tr eF + . ..) and if the n=O term is not to vanish then of course we must have b=l12• N=2 immediately follows since the remainder of (37) is already ind(ill+) correctly normalized. Thus 54 tr((-l)Fe-PH') = Ldet-t tsin~~}et[l +exp(~~ + icx )] (38) 47t ' where F = F µv dxµ Adxv, and the index of the twisted Dirac operator is 1 2 [sinhiRJ ind(iV'v+> =JM det ;_7t tr[ex~~~)J , 47t which is the result of the Atiyah-Singer index theorem (5). Note that the value that b must take is in the end consistent with the final 'Jl,'Tl integration being given as a trace - indeed Windey would not have made the assumption if it had not led to the correct answer! While the reason for this is not clear, it may be that after such manipulations as were performed the trace interpretation will always lead to the desired answer. Finally, note that the same problems would have arisen for bosonic c,c and 'Jl, fl. We would then have det~~~~t -F - icx), with the desired result def1 [1 - exp(F + icx)] = 1 + eiatr [ exp(F ) ] + ... Thus the problem has nothin~ to do with the definition of Grassmannian path integration. The problem does not arise, on the other hand, for second order operators or, what is equivalent, for first order operators acting on real fields, such as detPBc(gt 8aj3 - Raj3) the antisymmetry ofR here ensures that det(eR) = etrR = 1. We have thus derived, if not rigourously then consistently, the resuJt of the Atiyah-Singer index theorem for the Dirac and twisted Dirac operators using supersymmetric quantum mechanics and elementary path integral methods. In the process we have come to a better understanding of such methods and, though noting their drawbacks, have confirmed their strength. 55 References [1] M.F.Atiyah and I.M.Singer, Bull.Amer,Math.Soc. 69 (1963) 422, Ann.Math. 87 (1968) 485 and 546. [2] see, for example, L.Alvarez-Gaume, "An Introduction to Anomalies," lectures delivered at the International School on Mathematical Physics, July 85 (Erice) and at The Ecole Normale Summer Workshop, August 85, HUTP-85/A092. [3] E.Witten, "Fermion Quantum Numbers and Kaluza-Klein Theory", "Proceedings of the Shelter Island Conference", (MIT Press, 1985). [4] E.Witten, Nucl.Phys. B202 (1982) 253. [5] L.Alvarez-Gaume, Commun.Math.Phys. 90 (1983) 161, I.Phys.A 16 (1983) 4177. [6] P.Windey, Acta Physica Polonica B15 (1984) 435. D.Friedan and P.Windey, Nucl.Phys. B235 (1984) 395. [7] L.Girardello, C.Imbibo and S.Mukhi, Phys.Lett. 132B (1983) 69. [8] B.Zumino, "Supersymmetry and the Index Theorem," lecture given at the Shelter Island Conference, 1983, LBL-17972. I.Manes and B.Zumino, Nucl.Phys. B270 (1986) 651. [9] P.D.Iarvis and S.Twisk, Class.Quantum Grav. 4 (1987) 539. [10] P.Shanahan, "The Atiyah-Singer Index Theorem, an Introduction," Lecture Notes in Mathematics, vol.638, Springer-Verlag (1978). [11] see, for example, T.Eguchi, P.B.Gilkey and A.I.Hanson, Phys.Rep. 66 (1980) 213. I.L.Dupont, "Curvature and Characteristic Classes," Lecture Notes in Mathematics, vol.640, Springer-Verlag (1978). . [12] M.F.Atiyah, V.K.Patodi and I.M.Singer, Bull.London Math.Soc. 5 (1973) 229. M.F.Atiyah and I.M.Singer, Ann.Math. 93 (1971) 119. [13] S.M.Christensen and M.I.Duff, Nucl.Phys. B154 (1979) 301. [14] N.K.Nielsen, M.T.Grisaru, H.Romer and P.van Nieuwenhuizen, Nucl.Phys.B140 (1978) 477. [15] S.Adler, Phys.Rev. 177 (1969) 2426. R.Delbourgo and A.Salam, Phys.Lett. 40B (1972) 381. [16] K.Fujikawa, Phys.Rev.Lett. 42 (1979) 1195 [17] I.Wess and B.Zumino, Phys.Lett. 37B (1971) 95. W.A.Bardeen and B.Zumino, Nucl.Phys. B244 (1984) 421. [18] M.F.Atiyah and I.M.Singer, Proc.Nat.Acad.Sci.USA, 81 (1984) 2597. [19] M.F.Atiyah and F.Hirzebruch in "Essays on Topology and Related Topics," Springer-Verlag (1970). [20] see, for example, C.Itzykson and J.-B.Zuber, "Quantum Field Theory," McGraw-Hill (1980). L.Faddeev and A.Slavnov, "Gauge Fields: The Quantum Theory," Benjamin (1982). [21] L.P.Singh and F.Steiner, Phys.Lett. 166B (1986) 155. 56 [22] E.Getzler, Commun.Math.Phys. 92 (1983) 163. J.M.Bismut, Commun.Math.Phys. 98 (1985) 213. 57 Chapter 4 Grand Unification and Grassmannian Kaluza-Klein Theory 1. Preamble Kaluza-Klein theories [1] consider the forces of our four-dimensional world gravity and various Yang-Mills gauge theories - as being contained within the single gravitational force of a higher dimensional space. Recently [2] it was shown that the Einstein-Yang-Mills theory can also be found within a supergravity theory on a (4+N) dimensional superspace. That is the Kaluza-Klein idea also holds when the extra dimensions are Grassmannian. After reviewing briefly conventional Kaluza-Klein through the simplest (and original) such model, that based on a five-dimensional space-time, and then gravity on a superspace and the Grassmannian Kaluza-Klein ansatz for general Yang-Mills theories, we will consider in this chapter how the standard grand unified theories, the SU(5) and SO(lO) models, fit into the Grassmannian Kaluza-Klein framework [3]. 2. Conventional Kaluza-Klein Theory There are two levels at which Kaluza-Klein theory may be considered. At the level of the ansatz the form of the higher dimensional metric is restricted so that its four-dimensional field content is just that of gravity and Yang-Mills theory, or some slight generalization thereof, and so that the higher-dimensional Einstein-Hilbert action reduces to the Einstein-Yang-Mills action on substitution of the metric ansatz and integration over the extra dimensions. A more complete treatment is achieved when a specific form is chosen for the higher-dimensional space-time, such as (locally) M4xX, where X is a compact manifold whose length scale is of the order of the Planck length and which has the desired Yang-Mills gauge group as its symmetry group. Then the full higher dimensional metric may be expanded in the extra coordinates with . a suitable ansatz for the four-dimensional theory emerging naturally as the low-energy part and with, hopefully, the higher modes not contributing in any adverse manner. Let us consider the original five-dimensional Kaluza-Klein theory [4]. We will denote the co-ordinates of the five-dimensional space by (xµ,y), where xµ co ordinatize a four-dimensional space-time, and in order to distinguish the five dimensional metric, etc. from their four-dimensional counterparts we will write these quantities with a "' above. The five-dimensional Kaluza-Klein ansatz is " (gµv(x) - r Aµ (x)~(x) -KAµCx)) gMN(z) = (1) -lffiy(x) -1 or 2 · d~ = gµv(x)dxµ®dxv - (dy + KAµ(x)dx~®(dy + KA.y(x)dxv) (2) 2 where K =167tG, G being Newton's constant. This form for the metric is preserved under general co-ordinate 58 transformations of the xµ if gµv (x) transforms as a four-dimensional tensor, and under transformations of y of the form y -7 y + f(xµ) if~ r 1 Aµ(x) -7 Aµ(x) - K- a/(x), that is if Aµ(x) undergoes a U(l) transformation. It is convenient to work in the horizontal lift basis 9M of 1-forms, 5 0µ = dxµ, 8 = dy + KAµ(x)dxµ, and the dual basis DM of tangent vectors D =-2._ - KA (x)_Q_ D = _Q_ µ axµ µ oy' 5 oy rather than the bases dzM and '()/'()zM. Then [4] " (gµv(x) OJ MN (gµv (x) OJ gMN(z) = and ~ (z) = , Q -1 0 -1 the Levi-Civita connection is given by ~v(z) = r~v(x), fµv(z) = -; KFµy(x), where Fµv(x) = aµAy(x) - ovAµ(x), ~5 (z) = ~ KF~(x) = ~v(x), (3) with all other components vanishing, and the Ricci scalar is A 1 - _2 µv R(z) = R(x) - 4 ~F (x)Fµy(x) . Thus the Einstein-Hilbert action in five-dimensions ~ fd5 z~R(z) (4) reduces to the four-dimensional Einstein-Maxwell action 4 fd x Fi(~ R(x) - ! Fµv(x)Fµv(x)) (5) provided that the volume of the fifth dimension V=fdy= ~. The five-dimensional theory can be considered more fully if the space-time is taken to be M4xS 1, that is with the fifth coordinate y parametrizing the circle, which has, of course, U(l) as its symmetry group. Then gMN(z) may be expanded as a Fourier series in y. The zeroth order of this expansion will be like (1) except that it will also include a scalar field cp(x) allowing for the coordinate dependence of g55. The effective four-dimensional theory [5] which results from the substitution of this expansion in the action (4) involves the fields gµv(x), Aµ(x) and the Brans-Dicke field cp(x) together with an infinite tower of massive spin-two particles with masses 2rr.nN (ne Z+). In order to arrive at an effective four-dimensional theory involving matter 59 fields, these must be introduced into the original five-dimensional theory. The action for a scalar cl>(z) fdsz Ji~ g1'JN That is a massless scalar in five dimensions leads to an effetive four-dimensional theory involving a massless scalar not interacting with the electromagnetic field Aµ(x) together with an infinite tower of massive scalars which do interact with Aµ(x); the latter's masses and charges are 2nnN and 2nnK/V respectively. Requiring that the quantum of charge be the electronic charge e leads to the value V=2mc e for the "volume" of S1 the extra space, that is of the order of the Planck length, and thus the masses of the particles in the tower are of the order of the Planck mass. In order to discuss spinor fields,the vielbeins, eMA(z), and the spin connection, ~~(z), must be used. The viellbeins are such that " "A "B gMN(z) = T\AB~(z)~ (z) with the inverse vielbeins, EM A(z), satisfying "M "B B EA (z)~(z) =BA , etc. and the spin connection is given in terms of the Levi-Civita connection by " " " N "N " L '°MAB(z) = eAN(aM~ + r ML~). In the horizontal lift basis then, with the metric specified by the ansatz, "A (eµex(x) eM(z) = OJ (8) 0 1 /\ co µexp =coµexp /\ _1 µ v rosexp - 2 KEexFµvEp (9) /\ _1 EVF -" COµex5 - 2 K ex vµ - -coµSex with all other components vanishing. The five-dimensional action for a massless spinor \V(z) is Jdr 5z "e -;::'lf(Z)y .A" EA M (zhaMf + COMBC(z)l.." [r.B ' f ]) 'Jf(Z)" (10) . 8 60 The five-dimensional Dirac matrices, yA, can be realized in terms of the four dimensional ones, "f, as -.( = ('f,ir) where r = fy1fy, and with °o/(z) then being a four-dimensional Dirac spinor. Expanding 'i/(z) in a Fourier series 27tin y °o/(z) = f,e V 'If (n)(x) , and substituting (8) and (9) into the five-dimensional action (10) gives the eventual result d4 [·,..P.E µ(a - 21tin 1cA +.!.co [/ J x e~'I' -fr (n) 1 r a. µ v µ 8 µ~y '··?J) (11) - ....!.. E µF VF [- ~ "'1 21tin rJ~ 16 ~ I µv "( ' 1 ]r -v (n) A global chiral transformation of 'l'(n)(x) - i1t r 4 'l'(n)(x) ~ e 'l'(n)(x) eliminates r from (11), leaving the action as that of a massless, uncharged spinor together with, once again, an infinite tower of massive charged spinors. In addition all spinors have a Pauli coupling \jfFµv["f';yv]'I' to the electromagnetic fieldwith coupling constant K'./16. Note that none of these spinors is a candidate for the field of an electron, since here, as in the scalar case, they will only possess the electronic charge if their masses are of the order of the Planck mass, and in any case they possess anomalous magnetic moments. Light fermions are a major problem for realistic Kaluza-Klein theories. They can only emerge through the complicated topology of the extra space as zero modes of the appropriate Dirac operator. However, Witten [6] has shown that this mechanism can never produce a spectrum chirally asymmetric in its couplings to the Yang-Mills fields coming from the Kaluza-Klein mechanism. We discussed this briefly in Chapter 3, Section I.5. Other problems with Kaluza-Klein theories emerge when they are quantized. Taking the extra dimensions to be of the order of the Planck length leads to enormous Casimir energies and hence to an unacceptable value of the cosmological constant [7]. Also [5],[8] the combined effect of the infinite tower of spin-two particles becomes significant despite their large masses. On the other hand, taking the extra dimensions to be Grassmannian eliminates both of these problems (although, of course, other arise) as then the extra space need not be small in order to be unobservable and the expansion of fields in the extra coordinates is finite. 61 3. ·Supergravity We will now review the construction of supergravity [9] on a real ( 4+2N) dimensional superspace in the superbein formalism, preparatory to the consideration of the Grassmannian Kaluza-Klein ansatz itself. Let the superspace be locally co-ordinatized by zM = (xµ,~m), where µ=0, .. ,3, m=l, ... ,2N, xµ are commuting and ~m are Grassmannian. Then ~~=[MN]~~ where [MN]= {- 1 ifM=~, N=n [M] =[MM]. 1 otherwise ' On the superspace tangent vectors, T, and covectors, co, may be constructed. Locally T=~(z)_L a~ with {a~} a basis for the space of tangent vectors, they correspond to left differentia tion when acting on a function on superspace, and CO = dzMCJ\i(Z) where {d~} is the dual basis to {a:M}, ·so that co(T) = ~(z)CJ\i(z) = [M]coM(z)~(z) (12) In general, the commutativity of any function will be given by the indices which it carries; so that, e.g. , TM = (Tµ, Tm) with Tµ commuting and Tm Grassmannian. The order in which indices are contracted is therefore important. The general rule is that indices are contracted down, as in (12), with sign factors introduced should any other indices inteivene between the two which are contracted. From the vectors and covectors, tensors may be formed. A real (super pseudo-riemannian) metric tensor ds2, by means of which a correspondence between vectors and c~vectors is established (indices are lowered), is introduced. ds2 = d~gMN(z)[N]~ = ~dzMgMN(z) with gMN = [MN] gNM , and TN=~gMN. The inverse gMN of gMN , given by ~gNP = 8~ ' gPN~ = 8~ and satisfying ~N = [M][N][MN]~ , is used to raise indices, 62 Note that gMN is not a tensor, rather [M]gMN is, and contraction of its indices with others is an exception to our rule. Differential p-forms, 0, can also be defined through the exterior product of co-vectors (1-forms) d~AdzN = -[MN]dzN Ad~ . MP M n = dz A... Adz 1OM M (z) 1··· p The exterior derivative mapping p-forms into (p+ 1)-forms is defined by Mp M1 M a n ( dQ = dz A... Adz Adz M M M z) . az:·- i··· p The properties of these on superspace are fairly much those of those on ordinary space, although graded. The metric tensor ds2 may be written locally in terms of the flat metric ~AB=(~~ :J, where 11ab = (_~ ~) , by moving to an orthonormal frame {eA} of 1-forms, · -called-superbeins;- ·· - · ds2 = eAeB11BA (13) = d~e~azNeJ311BA = ~dzN[MN][AN]e~eJ11BA so gMN = [AN]e~eJ311BA . The properties UJ?.der conjugation of the superbeins must be established from (13) in accordance with the reality of the metric. The following conjugation rules are consistent ea* = ea , ea *= ea+N = t1 - , e..a-· = ea+N-~ = ea , ' where we have used a for l,... ,N and afor a+N, and we have now eA = (ea,ea,e7J The relation (13) is preserved under 0Sp(l,3/2N) rotations of the superbeins eA(z) ~ eB(z)Ls\z) , since 0Sp(l,3/2N) is the invariance group of11AB· Tensors under 0Sp(l,3/2N) may now be considered, and with the superbeins and their inverses we may change world indices M,N,etc. to frame indices A,B,etc. analogously to the conventional case. In order for the exterior derivatives of frame tensors to also transform tensorially a connection 1-form Cf> AB(z) must be introduced B M B Cf> A (z) = dz Cf>MA (z) and under a frame rotation 63 (15) so that, for example, ~-deA+eB- B A (16) B AB A cBA BA --7 de LB + e dLB + e c LB -e d~ =T1~A TA is called the torsion 2-form. The curvature 2-form RAB given by B B C B RA = d TlAB is talcen to be = ('la~ -e"' -e'") . The admissible co-ordinate transformations are restricted to just the ordinary 64 xµ ~ x'µ(x) em~em (19a) em~em and also local supersymmetry transformations xµ ~ ~µ - i(0cf~(x) - ~(x)cfe) m m m 0 ~ 0 - ~ (x) (19b) em~ em -~Ih(x) The admissible frame rotations are also restricted to those A Ls E OSp(l,3/4) of the form a L~ (x) A T. a(x) Ls (z) = 0 (19c) b L a.(x) where then L~ a, Lb a and Lb a are the same Lorentz transformation but in the vector, (1..,0) and (0,1..) = (.J-,0) representations respectively. 2 2 2 Finally the flat space metric is taken to be that one such that the supersymmetric differential operators D = (_Q__, l+ icf.eal, _Q_+ i0acf. ef>a_L) A axa aea aa axµ ()S. ab axµ a are orthonormal, i.e. M () DA =EA az11 giving the flat inverse superbeins EA M. From these the flat superbeins follow and hence the constraint on the torsion, namely ~ = (d0m dBm cflmm ,0,0) Ol T 65 4. The Grassmannian Kaluza-Klein Ansatz Let the (intended) Yang-Mills gauge group be G, with an N-dimensional unitary representation R. The crux of the Grassmannian Kaluza:-Klein theory is that 0(1,3)xSp(2N) c 0Sp(l,3/2N) and R(G)x R(G) c U(N)x U(N) c Sp(2N) where by R(G)xR*(G) is meant {R(g)xR*(g); ge G} and similarly for U(N)xU*(N). This is easily seen since, if a. -7 e~l a. ea -7 ebU a e ~ , b , where l~a.eO(l,3), UbaeU(N), then eBeAnAB = e~e0,,a.~ + ebeaOab - elJeaoab _ ~ a. 'Y I> _]} a * c d !J-=a c * d B A.,. -7 e e l~ la. T\&y + e e Ub Ua Ode - e e Ub Ua Ode= e e qAB The situation is thus similar to that of ordinary Kaluza-Klein theory when the Yang Mills group is the symmetry group of the extra space. Let the coordinate system of the superspace (xµ,sm) have the conjugation properties * m * m+N _:;n -m * m+N* m xµ = xµ , S = S = S , S = S = S m =l ,. . .,N Then the conjugation properties of objects with flat or curved space indices are the same. The Grassmannian Kaluza-Klein ansatz is [2] (from now on writing superfields with a " above to distinguish them from their four-dimensional counterparts) (20) or AA µ n a na e = (dx eµa.(x), p(dxv s Ay(x)n + ds on)) where A(x)n a is a matrix in the Lie algebra of R(G)xR*(G) and 2 p = exp(;2 s )' with s2 = SmS111nm and c a dimensionless constant. The component fields eµa.(x) and Aµ(x) are of course intended as the four-dimensional gravitational and gauge· fields respectively. The purpose of p will become clear later. The class of trapsformations which preserve the ansatz (20) consists of: general coordinate transformations of the xµ, xµ -7 x'µ(x), Sm -7 Sm (21) with then 66 (22) with then and "gauge transformations" when simu(ltan:ously a f)rame rotation with A_ 0~ 0 (23a) ~ - 0 uba(x) and a coordinate transformation m n m S -7 S Un (x) (23b) are made, where UeR(G)xR*(G), and provided that then 1 1 AµCx)-7 u- AµU - u- aµu (24) That is the transformations of the fields eµ 0 ) (gµv 0 ) ..:a , gMN = 0 23 . Pum P mn The torsion is constrained to A m a T = (0,0 Adp Orn) A which is the value of d~ when eµ a(x) = o; and Aµ (x) = 0. This together with the 67 condition AB = - [AB] µ ~ 2 m ~!: B dx roµa (x) - p 0 (Fa <:i)m (z) = ( ~ (25) A - dxµ(F µ S)aP where roµa~(x) is the Levi-Civita spin connection from eµa(x), and Fµv(x) is the Yang-Mills field strength tensor from Aµ(x) The curvature tensor and the Ricci scalar /\ A NA M AC A B R = [B]~ EA [ANJT\ RMNC can be found from (25). The result is [2] A 1 2i: µvi: R = R + "4 p 1:iFµvF 1:i . The Einstein-Hilbert action on the superspace is then 2 2 2 = 1.. Nt4z = 1.. Ns ep-2N[R + 2 (26) s a fd eR a Ja\ d lp4 sF s] where e= sdet(~A). The purpose of p can now be seen to be the provision of the necessary Sterms to saturate the s-integral. Performing this integral leads to the four dimensional action 4 ~ = Jd x e[~ R + 2~2 tr(FµvFv)J , (27) with Fµv (x) now taking values just in the Lie algebra of R(G), provided that .!.N(N+l) CX = 22N(-1)2 ~CN1C-2N-2 and 2 C = _ .!__ ( N-1 JN-1 4N N • Thus the unification of gravity and Yang-Mills theory within a (4+2N)-dimensional supergravity is achieved. 68 5. Matter Fields It is straightforward to write down the action for a general 0Sp(l,3/2N)- scalar superfield cp(z). It is jd2N+4z ~ ([MJgNM(a [p)*aNtp + B.2tp* lP + ~ To analyze this in terms of the component fields of i?(z), it will be convenient to write m Sn= S 'llmn so that ...iii m* Sm = - S and S = - Sm , m m-m and to work with (S ,Sm) rather than (S ,s ). Under a gauge transformation (23) m n m ...iii _fi. *m S -7 S Un (x), S -7 S (x) U_n 1 with U unitary, u- t = U*. So m n m * m -1 n S -7 S Un and Sm -7 Sn Un = Sn (U ) m (29) Expanding (p(z) m m 1 [p(z) = cp(x) + Sm cpm (x) + Sn 1 1 with cpm ••. m.n ••• n (x) ant~symmetric with respect to the interchange of any two mi 1 or any two nj. Since Cj)(z) is a scalar, under a gauge transformation 1 1 5 cpm ••• m.n ••• n (x) must transform in the ArR *®A R representation of G, where 1 AfR* denotes the anti-symmetrized Kronecker product of r R*s and similarly for A5R. When both r and s are non-zero these representations can be reduced by decomposing into traceless and trace parts, e.g. m m m2 m2 1 1 Sn S = (Sn S - 2N Bn S ) + 2N Bn S where 2 nm n S = 'llmnS S = 2Sn S • Also, if R(G) SU(N), then AfR* is isomorphic to AN-fR and A5R to AN-sR*. This equivalence can be seen through contracting with the completely anti-symmetric £-tensor, £ or em1···mN. m1···mN A typical irreducible term in the expansion of cp(z) is then, for r and s~N'2 , 1 ):2 tci: j: j:mr j: ml ) 111 ••• ns 1Crts+2t (':I ) ':ln&··':ln ':I ···':I - traces cpm ••• mr (x) 1 1 69 with cp itself being appropriately anti-symmetric and traceless. Substituting such a term into (28) as an ansatz for cp(z), the ~ 11 s and ~ms from cp alone cannot pair to saturate the ~-integral because of the tracelessness of cp. Instead the ~ms from cp and ~ 11 s from Cj>* pair, and vice versa. Thus after the integration Cj>*cp becomes cp t cp. Also Aµ(x), appearing in (28) through (~Aµ)mam, is automatically made to take its values in the representation carried by cp. So (28) becomes the conventional action 4 2 2 fd x e[ gµv((dµ-Aµ)cp) \av-Av)cp + m cptcp + A.(cptcp) ] (30) The m2 term here incorporates the T\mn(dmcp) t()ncp term of (28) as well as the m2 term. To the above the proviso should be added that if r+s+ 2t >Nf2, then the (fp*cp)2 term disappears due to a superfluity of ~ 11 s and ~ms , and that if r+s+ 2t > N then similarly the other, quadratic terms vanish. If a more complete ansatz were taken for cp(z) involving more than one irreducible term but all with different (r,s), then each term would lead to an action _such as (30) but there could also be interaction terms. These terms, however, could only , 2 come about through the A.(cp* rd2Nt4 A [A •,.ft µ a J: A Ila A A A A J' z e 'lf(z)1 r Ea ( µ - (~µ) n - roµ)'lf(z) + m'lf(Z)'lf(z) J (31) This action then decompqses into the conventional action for each irreducible component in the expansion of \j/(z) as for the scalar case. 70 6. Grand Unified Theories Using the foregoing it is now quite staightforward to consider cases of phenomenological interest. The main criterion that we apply is how easily can the required representations of matter fields be generated from the extra coordinates where L denotes the fact that these are left-handed spinors and superscript c denotes charge conjugation. The multiplets of scalar fields required for spontaneous symmetry breaking via the Higgs mechanism with the symmetry breaking route SU(5) ~ SU(3)xSU(2)xU(l) ~ SU(3)xU(l) are a 24 and a 5 respectively. The 5 also interacts with the fermions via a Yukawa coupling in order that they can acquire masses without breaking the chiral gauge symmetry of the model. Writing them as tensors based on the fundamental representation of SU(5) these fields are Hm(x) and "i:,nm(x) where "i:, is traceless. Extra scalar fields are required with Yukawa couplings to the fermions if the fermion masses are not to be related in an unrealistic manner. The possible representations are given by the form of the Y ukawa coupling efi'lf 2cp ' where 'l'i and 'l'z are left-handed spinors and cp is the scalar. This term must be overall an SU(5) singlet. Thus the possibilities are 5x5 = 10+15 5x10 = 5+45 lOxlO = 5+45+50 Of these the 10, the 15 and the 50 are ruled out on phenomenological grounds because 71 they do not have an electically neutral colour singlet component and thus, if they acquired a vacuum expectation value in order to give masses to the fermions at tree level, the they would also break the SU(3) colour group or the U(l) electromagnetism group. The 45 which remains can be written in tensorial form as K;\x) with K being antisymmetric with respect to interchange of m and n and also traceless. From the above we see that all matter field representations can be generated out of Grassmannian ~m and ~m transforming in the 5 and 5 representations of SU(5) respectively under gauge transformations. On the (4+ 10)-dimensional superspace the complete set of superfields required and their ansatzes in order to realize the model in the Grassmannian Kaluza-Klein framework are as follows. The superbein (32) with 4 p =exp(- 2~~ (~j ~2) . and Aµ taking its values in the 5x5 representation of the Lie algebra. The left-handed spinorial superfields " J: 1 J:m 1 J: m1J: mzi: ~ mn 'l'(x,..,) =!C.., 'l'm(x) + K3..., ..., ..., Em3mzm1nm X (x) and . x H(x,..,)" J: = -1 ~m H m (x) IC " 1( m 1 m 2) n :£(x.~) = r Sn s -10<>n s Lm (x) (34) and "J:_l( p 1 2 P ..P )!"n K(x,..,) - tc1 Sn Sm s -gS O>m Sn - on Sm) Kp (x)' with -:£ being commuting and -H and -K anticommuting. The superspace action is given by (26) 14 ~ fd z ~R. for the pure Einstein-Yang-Mills part, together with terms of the form (31) for \jl and -X , terms of the form (28) for H, I. and K and finally the Yukawa coupling terms 14 /\[ AC/\/\ /\C/\/\t /\C/\/\ /\C/\/\t ] Jd z e hlu'I' XH + Md'I' XH +Mu 'I' XK + Md 'I' XK +hermitian conjugates Note that the purpose of the extra term SSSEX in the superfield \j1 is to provide the Y ukawa couplings 72 and _m1mz m3m m4m5 Mu Emi1llz"l3m4ms Xc X Km The situation for the other popular grand unified group, SO(lO), is not so favourable. In an SO(lO) grand unified model the fermions of each generation, now including a right-handed neutrino, are gathered into a 16-dimensional multiplet of left handed spinors. Now the 16 is (one of) the spinor representations of SO(lO) and it cannot be generated from any of the other representations. Thus a Grassmannian Klauza-Klein theory based on SO(lO) must have 16 complex ~m. The Yukawa couplings possible for the 16 are given by examining 16x16 = (10+ 126)6+ 120a All of 10, 126 and 120 contain neutral colour singlet components and so are suitable. However, only the 120 which is the antisymmetric part of 16x16 can be generated from the ~m and ~m. In order to generate the 10 or the 126 a further 16 complex Grassmannian ~m must be introduced. The Grassmannian Kaluza-Klein theory is then based on a (4+64)-dimensional superspace. Fortunately this then suffices to generate the representations necessary for symmetry breaking, whether via ' 16 45 10 SO_(lO) -7 SU(5) -7 SU(3)xSU(2)xU(l) -7 SU(3)xU(l) or 54 - 45 SO(lO) -7 SU(4)xSU(2)LxSU(2)R -7 SU(3)xSU(2)LxSU(2)RxU(l) 16 10 -7 SU(3)xSU(2)xU(l) -7 SU(3)xU(l) In conclusion we can say that the SU(5) model is certainly more economical within a Grassmannian Kaluza:-Klein framework than an SO(lO) one. Of course, the Grassmannian Kaluza-Klein scheme as presented here is just at the level of the ansatz. Any analysis of the full spectrum of the theory would have the advantage over conventional Kaluza-Klein theory of there being only a finite number of modes. On the other hand, half of the modes are unphysical in their spin-statistics, so some mechanism mu~t be found to prevent them from interacting with the physical modes; the question of the spinor representations of OSp( 4/2N) also remains. 73 References [1] see for example H.C.Lee (ed.), "An Introduction to Kaluza-K.lein Theories," World Scientific (1984). [2] RDelbourgo and RB.Zhang, Phys.Lett. B 202 (1988) 296. [3] RDelbourgo, S.Twisk and RB.Zhang, Mod.Phys.Lett. A 3 (1988) 1073. [4] see D.J.Toms in [l]. [5] A.Salam, J.Strathdee, Ann.Phys. 141 (1982) 316. [6] E.Witten, "Fermion Quantum Numbers and Kaluza-Klein Theory," Proceedings of the Shelter Island Conference, MIT Press (1985). [7] S.D.Unwin, Phys.Lett. 103B (1981) 18. D.J.Toms, Phys.Lett. 129B (1983) 31. [8] M.J.Duff and D.J.Toms in "Quantum Gravity," ed. M.A.Markov and P.C.West, Plenum Press (1984). [9] see for example, I.Wess and I.Bagger, "Supersymmetry and Supergravity," Princeton (1983). [10] T.Eguchi, P.B.Gilkey and A.J.Hanson, Phys.Rep. 66 (1980) 22. [11] see for example, G.G.Ross, "Grand Unified Theories," Benjamin/Cummings (1985). 74 Chapter 5 Sp(2)-BRST Quantization 1. Introduction The BRST supersymmetry is of central importance in covariantly quantized gauge theories. The identities which it implies are necessary for renormalization [1] and the BRST operator is used in the physical state conditions which guarantee unitarity in the operator formalism [2]. Also the cohomology of the BRST operator is of relevance in the study of anomalies [3]. Yet in the conventional Faddeev-Popov approach to the construction of the gauge-fixed action, the BRST invariance only arises incidentally. Consequently a number of other approaches [4]-[8] have been developed which give a more central role to the BRST invariance. Fradkin and others [9] have introduced ghosts and subsequently the BRST invariance into Dirac's framework for the quantization of constrained systems. Methods of quantization based on BRST have recently been applied by many authors to the quantization of string theories [ 10]. There are further inadequacies of the Faddeev-Popov approach. It can only generate ghosts in pairs (ghost-antighost) and so can only cancel an even number of unphysical degrees of freedom. Also it only generates terms in the quantum lagrangian which are quadratic in the ghost fields. The former is unsatisfactory for antisymmetric tensor fields [11], for example - in four dimensions a second rank antisymmetric tensor field has six degrees of freedom but with the Kalb-Ramond action only one propagates. The latter will not always suffice for renormalization and unitarity [12], quartic ghost terms are possible and sometime also necessary. Another problem can arise if unitarity is being checked by the method of Kugo and Ojima [2], for this the square of the BRST operator must vanish without the use of the equations of motion requiring, in general, the presence of auxiliary fields; these are not given by the Faddeev-Popov technique. There is often in quantum lagrangians another symmetry, which is like the BRST symmetry with ghosts and anti-ghosts interchanged. It is known as the anti BRST symmetry [13], and the two together as the extended-BRST symmetry. The anti-BRST symmetry is not actually required for renormalization or unitarity and is not always present. Nevertheless it is remarkable that by requiring that the quantum theory have an extended-BRST symmetry the problems above do not arise. This requirement is made in a number of schemes [4],[6],[7] usually by formulating the gauge theory on a (4+2)-dimensional superspace coordinatized by zM = (xµ,0,S) or (xµ,em), m=5,6, with translation in the Grassmannian directions giving the extended-BRST transformations of the superfields of the theory. Bosonic fields Aµ .Jx) of the classical theory with some symmetry under interchange of 1 indices become superfields AM ... (x,8) with the corresponding graded symmetry. 1 This has been the approach used for tensorial fields in previous schemes based on a superspace, effectively it means that 0(1,3) representations are replaced by OSp(l,3/2) ones. Constraints are then imposed upon the component field content of the superfields in the "8-directions" through a curvature superfield or through coset 75 space dimensional reduction. The remaining component fields are then appropriate for the quantum theory corresponding to the original classical theory, with a translation invariant action for the superfields giving an action for the quantum theory which has an extended-BRST invariance. This method cannot be easily carried over to handle spinor fields as OSp(l,3/2) does not have suitable spinor representations. We give instead a less .ambitious approach [14] viewing the superspace more just as a device for the enforcing of the extended-BRST symmetry. By taking superfields only in 0(1,3) representations we avoid the need to impose constraints and the problems with spinor fields, while still getting all the results of the more ambitious schemes. We maintain an Sp(2) symmetry in our formulation· as this gives a more compact form for the extended-BRST transformations and provides a more general framework for ghosts than the ghost-antighost form. Before considering this we will review the Faddeev Popov approach for gauge theories. 2. Faddeev-Popov Quantization .will BRST The presence of a local gauge invariance is a problem in the quantization of a classical field theory in that it implies that the operator in the quadratic approximation to the lagrangian for the gauge fields cannot be inverted to give the gauge field's propagator. This reflects the fact that the equations derived from the principle of least action are nc;>t all equations of motion, some of them are constraints. In principle, in the non-interacting case these constraint equations could be solved and used to eliminate the non-propagating degrees of freedom of the gauge field. However, the remaining degrees of freedom will not, in general, carry a representation of the Lorentz group so that the manifest covariance of the theory will be lost. In any case, when interactions are present the constraints will not usually be invertible. The alternative to this approach of restricting the phase space is to add a gauge-fixing term to the lagrangian to break the gauge invariance so that the propagators can be found. The space of physical states corrsponding to incoming, and hopefully outgoing, particles in a scattering process can then be restricted to those which propagated in the gauge invariant theory. This suffices for electromagnetism. However, for self-interacting theories such as non-abelian gauge theories and gravity the presence of negative norms in the space of unphysical states leads to violations of unitarity. It was realized early [15] that, in the context of Feynman diagrams, unitarity could be restored if extra fictitious fields, called ghosts, were added. These fields by having spin-statistics opposite to those prescibed by the spin-statistics theorem serve to restore unitarity. The understanding was that the enlargement of the phase space to include the unphysical modes of the gauge field must be compensated for by the addition of the same number of ghost degrees of freedom. This idea has been put on a sound footing by Fradkin and others [9]. The arguments above still leave 76 undetermined the ghost lagrangian, unless it is still to be found by trial and error from the Feynman diagrams. Faddeev-Popov [16], working in the path integral approach, resolved this difficulty to some extent. Their approach is briefly described below for a Yang-Mills gauge theory. Classically, the Yang-Mills theory with a gauge group G is described by the gauge field a Aµ (x) =Aµ (x)ta taking its values in the Lie algebra of G, with the curvature Fµv(x) = aµAv(x) - ()vAµ(x) + [Aµ(x),Av(x)] and, when the Lie algebra is taken in some representation such that the generators ta are normalized so that with action 4 Jd x tr(~ Fµ/x)Fv(x)). The action is invariant under the gauge transformations Aµ(x) ~ A!(x) = g-1(x)AµCx)g(x) + g-1(x)aµg(x), (1) where g(x) takes its values in the corresponding representation of G, since then Fµv(x) ~ g-1(x)Fµv(x)g(x). Infinitesimally g(x) = eA(x) = 1 + A(x) , where A(x) = A a(x)ta, and Aµ(x) ~ Aµ(x) + DµA(x) = Aµ (x) + aµA(x) + [Aµ ~x),A(x)] (2) In the quantized theory the generating functional is given by the path integral 4 Z[J] = fDAµ(x) exp(ifd x tr(~ Fµv(x)Fv(x)-2J(x)Aµ(x))) (3) where the prime denotes that the integration should be carried out only,over a space of physically distinct or gauge-inequivalent Aµ(x) given by some gauge condition. For example, the space of Aµ(x) given by the covariant condition aµ Aµ(x) = A.(x), (4) for some Lie algebra valued A.(x). While Z[Jµ] is different for different choices of gauge condition, physical scattering amplitudes are not. The path integral (3) with the condition (4) can be written as an integral over all Aµ(x) 4 2 Z[J] = fDAµ(x) S(aµAµ(x) - A.(x)) A( Aµ(x)) exp(ifd x tr(~ F - 2J-A)), (5) through the insertion of a 8-function and also a A(A), where (3) and (5) together serve to define A(A). It can then be shown that 77 A(A) = det(aµDµ) where the operator o·D arises from µ g µ µ a Aµ(x) - a Aµ(x) =a DµA(x), in which g = exp(A(x)). A(A) is known as the Faddev-Popov determinant. Since physical quantities are not affected by the choice of A.(x), it may be integrated over with a gaussian weight 2 fDA.(x) exp(ifd\ tr(~ A. )) in order to eliminate the o-function from the expression for Z[Jµ]. Finally the Faddeev-Popov determinant can be written as the gaussian path integral over a new set of Lie algebra valued fields ro(x), ro(x) A(A) = Jnro(x)Dro(x) exp(ifd\ tr(-2roa.nro)) where ro(x) and ro(x) must be Grassmannian, in order for the integral to give A(A) and not A-1(A), and thus violate the spin-statistics theorem. (5) now beco~es 2 Z[J] = fDAµ(x)Dro(x)Dro(x) exp(ifd4x tr(~ F2+ ~ (o·A) - 2rooµDµro - 2JAµ) }<6) with the quantum action 2 2 . S[Aµ(x),ro(x),ro(x)] = tr(; F + ~ (o·A) - 2rooµDµro) (7) containing gauge fixing and ghost terms. The fact that the quantum action is not itself invariant under gauge transformations considerably complicated the renormalization program until it was observed [l] that the quantum action retains a global symmetry, the BRST symmetry, given exactly by the transformations BE~ =EDµro o ro=-.!.e{roro} (8) E 2 ' o ro =-.!.ea Aµ E 2 µ ' where E is an anticommuting parameter. These transformations are nilpotent when the equations of motion are taken into account. To ensure nilpotency off-shell an auxiliary field B(x) must be introduced. 2 2 S[Aµ'ro,ro,B] = Jd4x tr(; F - ~ B + 2BoµAµ - 2roaµDµro) (9) is invariant under the transformations given by BE= ES' 78 sAµ =Dµro sro = - l {ro,ro} (10) 2 sro=B sB =0 where s, the BRST operator, is linear, graded-Leibniz, commutes with aµ and is nilpotent s2 = 0. For such a theory, invariant under a nilpotent BRST transformation, Ku go and Ojima [2] showed that unitarity could be achieved in the framework of canonical quantization with the space of physical states restricted to a certain subspace of those which are annihilated by the BRST operator. The anti-BRST symmetry of the action (9) is given by sAµ =Dµro sro = - {ro,ro} - B (11) sro = - l { ro,ro} 2 SB= - [ro,B] It is also nilpotent, -2S - 0 . For a different gauge fixing the anti-BRST transformation will talce a different form. For example, in the axial gauge, the action 2 2 S[Aµoro,ro,B] = fd\ ~~ F - ~ B + Bnµ Aµ - 2ron~µro) is invariant under (10) but the anti-BRST transformations are given by sAµ = Dµro sro=B sro = -l {ro,ro} 2 SB =0 Note that quartic ghost terms of the form 4 2 fd x tr( {ro,ro}) are acceptably both on the grounds of renormalizability and of BRST invariance, although not generated in the Faddeev-Popov argument and not actually required for the renormalization of (9). The Faddeev-Popov result might be summarized using the BRST transformation in the following way. Given a gauge theory with gauge field A(x) and classical action invariant under the transformation 79 A(x) ~ A(x,A\x)) , the quantum theory consists of the gauge field A(x), ghosts ro 3 (x), antighosts ro 3 (x) and auxiliary fields B3 (x). The BRST transformations are given by =ES 0E 3 oEA = A(x;Ero ) - A(x) (12) 3 3 sro = B and s2 = 0 sro3 follows from 8EA and s2 = 0 and the requirement that s be a linear, graded- Leibniz operator commuting with aµ- The BRST-invariant quantum action is Lc(A) + s(ro\f (A) + BJ)) (13) fd4x ( 3 2~ where f3 (A) are chosen to give a suitable gauge fixing. Once again, this cannot lead to quartic ghost terms which are necessary for renormalization in non-linear gauges in Yang-Mills theories and unitarity in supergravity [7],[12]. The other inadequacy mentioned in the introduction can be seen for the antisymmetric tensor field:- In four-dimensions, the second-rank antisymmetric tensor field theory is based around the free action 4 ( 1 v pcr )2 _ J 4 ( 1 µv µ pv ) Jd x 2 Eµvpcra A (x) - d x 2 A (x) Aµ/x) +a Aµ/x)aPA (x) (14) which is invariant under Aµ/x) ~ Aµ/x) + aµAy(x) - avAµ(x) Applying (12) sAµv =aµO\. - avroµ SO)µ= Bµ (15) sBµ = 0 and sroµ must satisfy (16) The quantum action (12) is 4 ( _µy µy 1 µ d x Lc(A) + ro a (aµO\. - avroµ) + B a Aµv + a B Bµ) (17) J 2 where we have taken fµ (A) = av Aµv . This action has secondary invariances roµ ~ roµ + aµA (18) roµ ~ roµ + aµA which could be fixed through a further application of (12) and (13), introducing secondary ghosts, now with physical spin statistics, and altering (15) to allow for these. However, introducing two pairs of secondary ghosts for (18) gives overall two 80 (6-8+4) physical degrees of freedom, while the correct result is one - that is there should only be three secondary ghosts. We will see below that this multiplet can be understood in an Sp(2) framework, rather than in terms of ghosts and antighosts. 3. SpC2)-BRST The prescription [14] that we give here as an alternative to the one (12) resulting from the Faddeev-Popov method is based on the requirement that the set of quantum fields carry a representation of the extended-BRST algebra, which is isomorphic to the two-dimensional abelian superalgebra T(2), generated by sm (m = 1,2) with {sm,Sn} =0, the form of the transformations of the original classical fields coming from their gauge transformations and the algebra giving the transformations of the remaining quantum fields. We introduce this algebra by means of a superfield construction based on a superspace coordinatized by (xµ,em), with metric 1lµv on the commuting part and Eillll (e12 = -1) on the anticommuting part. The generators of translations in em form the superalgebra T(2), and their action on the superfields we take to be equivalent to the action of the extended-BRST transformations on the component fields. The invariance group of emn is Sp(2), so that working covariantly leads to an Sp(2) symmetry between the two BRST transformations, which together we will call the Sp(2)-BRST transformations. In all, the set of quantum fields will carry a representation of ( O(l ,3)xSp(2) )AT( 4/2). For a gauge theory as before involving gauge fields A(x) and also matter fields 'Jf(x), with classical action 4 fd x Lc(A(x),'Jf(x)) (19) invariant under the transformations A(x) ~ A(x;Aa(x)) (20) 'Jf(X) ~ 'Jf(x;Aa(x)) we form the superfields A(x,8) and 'Jf (x,8) by making super-local gauge transformations A(x,8) = A(x;A a(x,8)) (21) 'Jf(x,8) = 'Jf(x,A \x,8)) with 2 A a(x,8) = Smro!(x) +; 8 ba(x) (22) 2 m m n • where 0 = 0 Sm = £mn0 0 . A 0-mdependent part of A a(x,0) need not be considered as it may be factored out and absorbed into an ordinary gauge 81 transformation of the classical fields A(x), '\jl(x). We have then A(x,e) 1 A(x) 0 = 0 = = O '\jl(x,e)l9 = 'l'(X) Under an infinitesimal translation em~ em +Em a superfield F(x,e) has variation BEF(x,e) = F(x,e+e) - F(x,e) to the first order in e =~ am F(x,e) (23) ae which we interpret as being (24) where emsm act on the component fields of F(x,e) giving their variantions under the Sp(2)-BRST transformations with parameters em. Then 2 F(x,e) = F(x) + em smF(x) + l. e l. EmnsmsnF(x) . (25) 2 2 ' It also follows that {sm,snl = 0, [sm,aµ] = 0 and that sm is linear and satisfies a graded Leibniz rule, i.e. sm(F(x)G(x)) = (smF(x))G(x) ± F(x)(smG(x)) according as F(x) is commuting or anticommuting. Comparing (25) with the superfields A(x,e) and '\jl(x,e) given by (21) immediately gives the expressions for the Sp(2)-BRST transformations of A(x) and '\jl(x). It also gives expressions involving the Sp(2)-BRST transformations of coma(x) and ba(x). These latter can be solved, with any local freedom in the solution indicative of a secondary gauge invariance and automatically generating new independent fields - the secondary ghosts. This process is continued until the Sp(2)-BRST algebra closes on the space of fields. To each quantum field there will then be a corresponding superfield of the form (25). The quantum action for the gauge theory may now be formed. It consists of the original classical action (19) together with gauge-fixing and ghost terms. The gauge invariance of (19) ensures its invariance under the Sp(2)-BRST transformations 4 fd x LiA(x),'l'(x)) = Jd\ Lc(A(x,e),'\jl(x,e)) = fd4x LiA(x,e+e),'Jl(x,e+e)) The form of the remaining terms must be determined separately for each theory in accordance with the requirements of renormalizability and unitarity. The former requires that the canonical dimension of the terms in the lagrangian density be less than or equal to four without the introduction of dimensionful constants, and the latter that the action be hermitian. The hermiticity and dimensionality of coma(x), ba(x), etc. 82 and of em and sm must be determined in accordance with these requirements. Of course, there must also be at least one BRST-like invariance of the quantum action. The full Sp(2)-BRST invariance can be assured for terms of the form 2 Jd 0 f(A(x,0), ... ) = ~ Emnsmsnf(A(x), ... ) due to the invariance of the 0-integral under 0-translations or due to the fact that {sm,sn} = 0. Sp(2)-BRST invariant terms of the form smgn(A(x),rop(x), ... ) , where gm is an Sp(2) vector of the appropriate canonical dimension, with must then satisfy 1 2 -0 -S2 gm - ' but with gm "::/:. smf for any f, may also be possible, in general. For the cases of the Yang-Mills field, the anti~ymmetric tensor field and the Rarita-Schwinger field, whose gauge transformations are of the form BAA(x) = "iJA + ... and which are considered below, the general form 4 Jd x (Lc(A(x)) + ~ Emnsmsn(~ 1 (A(x),A(x)) + ~2Epq(rop(x),roq(x)) + ... )) , (26) where ~ 1 , ~2 are dimensionless, (A(x),A(x)) is an inner product which breaks the gauge invariance and (ro(x),ro(x)) is a suitable bilinear, gives correct quantum actions. Here the canonical dimensions are [s] = 2 -[A], [ro] = 1, [b] = 2, etc. and the hermiticity assignments are at a 01t -- 02 ro1 = - ro2 , , etc., order being reversed under hermitian conjugation. 4. Yang-Mills Applying our prescription to the pure Yang-Mills case, we form the superfield Aµ(x,0) = Aµ(x;A(x,8)) _ e-A(x,8)A ( ) A(x,0) -A(x,8)'.'.\ A(x,0) - µ x e +e uµe • m 2 a where A(x,8) = 0 rom(x) + 1 0 b(x) and A= A ta , etc. 2 Then 83 2 Aµ(x,0) = (1 -A+; A ) Aµ (1 +A+; Al+ (1- A) (oµA +;()µAA+; A()µA) = Aµ(x) + DµA + ; [DµA,A] 2 = Aµ(x) + 0mDµrom(x) +; 0 (Dµb(x) +; emn{Dµcom(x),ron(x)}) (27) using e men - --e1 mne2 2 we have (28) and that ; emnsmsnAµCx) = Dµb(x) +; emn{Dµcom(x),ron(x)} (29) Now the algebra {sm,sn} = 0 implies that smsn is antisymmetric SmSn = Emn ; EpqSpSq (30) so that smsnAµ(x) = Erun; ~spsqAµ(x) = Erun(Dµb(x) +; epq{Dµcop(x),coq(x)}) but also, using (28) and [sm,aµ] = 0, SmsnAµ(x) = smDµron(x) = smaµcon(x) + sm[Aµ(x),ron(x)] = D/smron(x)) + {Dµcom(x),con(x)} = Dµ(Smffin(i<.)) + ; Dµ{ rom(X),ron(x)} + ; ErunEpq{Dµcop(x),coq(x)} where the anticommutator term has been split into symmetric and antisymmetric parts with respect to mn. Comparison yields Smffin(x) = Erunb(x) - ; { rom(x),con(x)} (31) To find smb(x), note that so that from (29) and then Dµ(smb(x)) = - [Dµrom(x),b(x)] - ; enpsm {Dµron(x),cop(x)} which using (28) and (31) eventually yields Smb(x) = ; (b(x),Olm(x)] + 1 Epn( {COm(x),COn(x)} ,COp(x)] (32) 12 The quantum action taken in the form of (26) is 84 2 fd4x tr[; F +; emnsmsn(~ 1 Aµ(x)Aµ(x) + ~epqroq(x)rop(x))J = fd4x tr [; F + ~ 1 (Aµ(x)emnsmsnAµ(x) + emnsmAµ(X)SnAµ(x)) + ~(epqroq(x)EmnSm8nCllp(X) + EpqEmn8nroq(x~smrop(x)) J 4 2 = fd x tr [ ~ F + S1(2Aµ(x)aµb(x) + emnoµrom(x)aµ ron(x)) 2 + ~2 (2b (x) - ! emnepq{ rom(x),rop(x)} {ron(x),roq(x)}) J (33) The Sp(2)-BRST transformations become the familiar BRST and anti-BRST transformations (10) and (11) if we write rom(x) = (ro(x),ro(x)) , sm = (s,s) and b(x) = B(x) + ~ {ro(x),ro(x)} In terms of these fields (33) is 2 fd 4x tr[ ~ F + ~ 1 (- 2BaµAµ + 2roaµDµro) 2 2 + si2B + 2B {ro,ro} + j ({ ro,ro }) - ~ { ro,ro }{ ro,ro})] (34) which for 1;1 = -1, 1;2 = 0 is just (9) in the Landau gauge a~ 00 • The extra terms which accompany B2 are aconsequence of the Sp(2) invariance of our lagrangian. The general form of (33) and (34) is equivalent to those arrived at in [5],[6], where has been shown to lead to the correct quantum theory. Our scheme is different from these other schemes in that we do not see the superspace (xµ,em) as anything more than a device for the imposition of the extended-BRST symmetry. That is we do not attempt a unified treatment of the commuting and anticommuting directions. The other schemes by contrast write down a superspace version of Yang-Mills theory involving a superfield AM(x,0) = (Aµ(x,0),Am(x,0)) By imposing the condition that admissible AM(x,0) are just (Aµ(x),O) up to a super local gauge transformation, they arrive at AM(x,0) = (Aµ(x;A(x,0)),rom(x)+0mb(x)) (35) The superfield Aµ(x,0) is now just the same as ours. No real unification is achieved in these schemes because their own condition (35) denies any equivalence between the xµ and em directions - it is not preserved under a general 0Sp(4/2) transformation. As well, that approach cannot easily be applied to spinorial matter fields. Generalizing spinors to the superspace would require them to carry a representation of the Clifford superalgebra 85 but the sector ['f1;f] = 2emn has only infinite dimensional representations (as we pointed out in Chapter 4, this is just the simple harmonic oscillator [a,a] = 1) - related to the fact that 0Sp(4/2) has no finite dimensional spinor representations of non-vanishing superdimension [6],[17]. For us the incorporation of matter fields '!'(x), transforming under gauge trans-formatjons as -Aa(x)t 'l'(x) ~ e \v(x) in some representation of the ta, is straightforward whether they are Lorentz scalars or spinors. We form the superfield -A(x,8) ( ) 'I'( X, 8) = e 'I' X 2 = 'l'(x) - A(x,8)'1'(x) + lA (x,8)'1'(x) 2 -··-·---· -= 'Jl(Xj--0mrom(X) +; e2<; EmnCOm(X)COn(x) - b(x))'!'(X) and thus find that the Sp(2)-BRST transformation of 'l'(x) is Sm'l'(X) = - rom(x)\j/(x) (36) This together with (28) smAµCx) = Dµffim(x) gives an invariance of the gauge-invariant matter lagrangian. 5. Anti-symmetric Tensor From the antisymmetric tensor field Aµv(x) with classic~l action (14) we fomi the superfield Aµv(x,8) = Aµv(x) + aµ~(x,8) - avAµ (x,0) m 1 2 where Aµ(x,8) = 8 romµ(x) + 2 0 bµ(x) Thus finding that smAµv(x) = aµromv(x)- avromµ(x) (37) sm(aµronv(x) - avronµ(x)) = Enm(aµbv(x) - avbµ(x)) and sm(aµbv(x) - avbµ(x)) = 0 Solving, we generate the fields romn(x) and bm(x). smronµ(x) = Enmbµ(x) + aµromnCx) (38) smbµ(x) = aµbm(x) (39) 86 Any antisymmetric part of comn(x) may be absorbed into emnbµ(x) so that we talce it to' be symmetric. To find the Sp(2)-BRST transformations of comn(x) and bm(x) we use the antisymmetry of spsm. SpSmCOnµ(x) = Erundµbp(x) + dµSpCOmn(x) which must be, for some dn(x), = 2Epmdµdn(x) Comparing parts antisymmetric or symmetric in nm, we have ~(x) = bp(x) and (40) Finally, SqSpSmCOnµ (x) = 0 ' since it is completely antisymmetric in qpm, implying that Sqbp(X) = 0. The quantum action in the form (26) is 4 fd x [ Lc(A(x)) + ~ EmnSm8n(s1Aµ\x)Aµv(x) + ~comµ(x)comµ(x) + ~romn(~)romn(x))] which is · 4 fd x [Lc(A(x)) + 4s1bµ(x)av Aµv(x) + 2s1aµ comv(x)(avcomµ(x) - aµ'°mv(x)) + 2s bµ(x)bµ(x) + ; aµ comn(x)aµcomn(x) - 6s bm(x)bm(x) (41) 2 2 3 J Here Aµv(x), '°mµ(x) and romn(x) propagate, with the physical degrees of freedom being 6 - 8 + 3, = 1 as it should, with the correct number (three) of secondary ghosts coming in through a repre.sentation of Sp(2). 6. Rarita-Schwinger This case has been treated [18] in an 0Sp(4/2) framework using an infinite cµmensional representation of the Clifford superalgebra. However, it was found that a correct result could not be attained. On the other hand, our method although simple minded does handle this case satisfactorily. Here the gauge field is a spinor-vector 'l'µ(x) with classical action 4 4 1 fd x Lc('l'µ(x)) = fd x i'lfµ(x)Yµ'°lr av'l'p(x) 4 = fd x i'lfµ(x)(Ttµvjl'- (dµ "l + avf-y-;J{))'Jly{x) invariant under the gauge transformation 'lfµ(x) ~ 'lfµ(x) + i()µA(x) , with A(x) a spinor. 87 2 By taking A(x,8) = 8mcm(x) + .!..8 B(x), we form the superfield 2 'l'µ(x,8) = 'l'µ(x) + iaµA(x,8) 2 = 'l'µ(x) + 8midµcm(x) + ; 8 i()µB(x) (42) Then we have Sm'lfµ(x) = idµCm(x) SnCm(X) = EronB(x) (43) SnB(x) = 0 We take and (cm(x),cn(x)) = cm(x'ifcnCx) where a is required to ensure that this term has the correct dimension, then the form (26) gives as the quantum action 4 Jd x [Lc('l'µ(x)) +; Emnsnsm(~ 1 '1'µ(x)y•{'l'v(x) + ~2epqcq(x)/cp(x))] = Jd4x [ Lc('l'µ(x)) + ~ 1 (emncm(x) cn(x) + i'l'µ(x)yi) B(x) - iB(x)/ Y'l'µ(x)) + ~2B(x)/B(x)] This is the quadratic part of the lagrangian that is shown in[l 9] to lead to a unitary quantum theory for the gravitino part of supergravity. Conclusion In this chapter we have discussed a simple prescription for quantizing gauge theories based upon two BRST symmetries related by an Sp(2) transformation. The Sp(2) symmetry leads to a more compact description of the extended-ERST symmetry than does the BRST-anti-BRST approach and provides some rationale for the occurrence of ghosts other than in ghost-antighost pairs. By not attempting to unify the BRST with the space-time symmetries the problems which a method based on 0Sp(4/2) encounters with spinorial fields have been avoided. 88 References [1] C.Becchi, A.Rouet and R.Stora, Ann.Phys. 98 (1976) 287. I.V.Tyutin, Report PIAN 39 (1975). [2] T.Kugo and I.Ojima, Prog.Theor.Phys.Supp. 66 (1979) 1. [3] L.Bonora and P,Cotta-Ramusino, C.M.P. 87 (1983) 589. [4] S.Ferrara, O.Piguet and M.Schweda, Nucl.Phys. B119 (1977) 493. K.Fujikawa, Prog.Theor.Phys. 59 (1978) 2045. [5] L.Bonora and M.Tonin, Phys.Lett. 98B (1981) 48. L.Bonora, P.Pasti and M.Tonin, Ann.Phys. 144 (1982), l~. [6] R.Delbourgo and P.D.Jarvis, J.Phys.A 15 (1982) 611. R.Delbourgo, R.J.Fanner and P.D.Jarvis, Phys.Lett. 123B (1983) 319. [7] L.Baulieu, Phys.Rep. 129 (1985) 1. [8] D.Nemeschansky, C.Preitschopf and M.Weinstein, Ann.Phys. 183 (1988) 226. [9] E.S.Fradkin and G.A.Vilkovisky, Phys.Lett. B55 (1975) 224. see for a review: M.Henneaux, Phys.Rep. 126 (1985) 1. [10] for example, L.Baulieu, C.Becchi and R.Stora, Phys.Lett. 180B (1986) 55. J.P.Ader and J.C.Wallet, Phys.Lett. 192B (1987) 103. W.Siegel and B.Zwiebach, Nucl.Phys. B288 (1987) 332. R.Delbourgo, P.D.Jarvis, G.Thompson and R.B.Zhang, Mod.Phys.Lett. A 3 (1988) 303. [11] P.K.Townsend, )?hys.Lett. 88B (1979) 97. T.Kimura, Prog.Theor.Phys. 64 (1980) 357. P.A.Marchetti and M.Tonin, NuovoCim. 63A (1981) 459i [12] G.Curci and R.Ferrara, NuovoCim. 32A (1976) 151. P.van Nieuwenhuizen, G.Sterman and P.K.Townsend, Phys.Rev.D 17 (1978) 3179. N.K.Nielsen, Nucl.Phys. B140 (1978) 499. R.E.Kallosh, Nucl.Phys. B141 (1978) 141. [13] G.Curci and R.Ferrari, Phys.Lett. B63 (1976) 91. I.Ojima, Prog.Theor.Phys. 64 (1980) 625. [14] S.Twisk and RB.Zhang, Mod.Phys.Lett. A 3 (1988) 1169. [15] R.P.Feynman, Acta.Phys.Pol. 26 (1962) 697. B.S.DeWitt, Phys.Rev. 162 (1967) 1195,1239. [16] L.D.Faddeev and V.N.Popov, Phys.Lett. 25B (1969) 29. see, for example, C.Itzykson and.J.-B.Zuber, "Quantum Field Theory" McGraw-Hill (1980). [17] J.Thierry-Mieg, Phys.Lett. 129B (1983) 36. [18] J.A.Henderson, PhD Thesis, Uni. of Tasmania (1987). [19] H.Hata and T.Kugo, Nucl.Phys. B158 (1979) 357. 89 Chapter 6 Massive Yang-Mills Theory: Renormalizability vs Unitarity 1. Preamble The Higgs mechanism is the standard device by which massive non-abelian gauge fields are considered within a renormalizable and unitary theory. Yet, until and unless a Higgs particle is observed experimentally, this method must always be open to some doubt. The search for such a particle is of course made difficult by the fact that the Higgs method does not of itself determine the particle's mass nor its coupling constants in terms of other known constants - although constraints upon the values possible have been obtained indirectly. Another problem with the Higgs mechanism arises within the context of grand unified theories - the so-called hierarchy problem. Recently another approach was suggested [l] as a possible alternative to the Higgs method. It is based upon the non-abelian generalization of the Stueckelberg model. Unfortunately, it was soon discovered [2],[3] that although this approach leads to a renormalizable theory it is only so at the expense of unitarity - whereas ordinary massive non-abelian vector theory might be thought of as unitary but unrenormalizable. We shall in this chapter demonstrate the conflicting nature of the two requirements. 2. Renormalizablity Qf Massive Gau@ Theories and ~ Stueckelberg Model We begin with a brief review of massive electrodynamics. This theory is renormalizable even though naive power counting would lead one to the opposite conclusion. The massive Lagrangian, without matter fields, is _µv , 2 µ L8 = - ! Fµv(x)t< (x) + ~ Aµ(x)A (x) (1) where Fµv(x) = aµAv(x) - avAµ(x). The mass term is not invariant under the gauge transformation Aµ (x) ~ Aµ (x) + dµA(x) and thus the propagator may be found straightaway. It is ~~ 1lµv--2- m (2) 2 2 k - m +ie and is of order 1 at high momenta, leading to the expectation that when interactions are added the result will be unrenormalizable by power counting. This is, however, not the case if the coupling is to a conserved current [4], such as in spinor electrodynamics where L = L(Aµ(x)) + g{(x)Aµ(x) + Lro('l'(x),'!'(x)) (3) / (x) ='l'(x)y'!f(x) Lro('1'(X),'1'(X)) = i'\j/(x)ydµ '!f(X) - M\ji(X)'!f(X) and for which 90 gt(x)Aµ(x) + ~(\jf(x),\jf(x)) is invariant under a gauge transformation, Aµ(x) --t Aµ(x) + aµA(x) \jf(x) -7 eigA(x)\jf(x) _( ) -igA(x)_( ) \jfX --te \jfX. To see that such a theory is renormalizable we use the Stueckelberg formulation which restores gauge invariance. It is obtained by introducing the Stueckelberg field cp(x) through the gauge transformation Aµ(x) --t Aµ(x) - ~ aµcp(x) giving the Lagrangian 2 2 2 L = -.!.F + m (A -..!..a + gtA + Lm (4) 4 2 µ m µ cp) µ which is invariant under BAµ =aµA Bcp = mA "A 'I' -7 eig 'I' 'l'-7 ei~ and so must be gauge fixed before propagators can be found. Consider first of all the gauge fixing given by the function f(A) = a. A. To (4) is added L f = s(ro(a·A + -1 B)) g 2a where s is the BRST transformation scp = mro sAµ = aµro, etc. and sro=B SOO= 0 (5) sB =0, s2 = 0 · ' i.e. 1 2 L f =Ba.A+ +--B g coo:o 2a The propagator for Aµ(x) may now be found and it is just the standard propagator 91 1-a l)ikv 11µv - """"(l 2 2 k -m /a+ ie 2 2 k - m + ie which behaves as lfk.2 for large k, thus ensuring renormalizability. Another gauge choice that may be considered is L r = s(ro(q> +-1 B)) g 2a 2 = Bq> + mroro +-1-B 2a Since the lagrangians for the two cases differ only by a BRST invariant term the S matrices which each give will coincide [5]; but for the latter, as ex -7 oo, q> = 0 is enforced and the theory is manifestly the original one (4). Thus this theory must be renormalizable. The non-abelian case is, however, not so favourable. Here the massive lagrangian is (6) where Fµv = aµAy - avAµ + g[Aµ>Ay] . a a a Aµ= Aµta, Fµv = Fµvt [ta,~] = fabctc' tr(ta\) = - ~ Bab The mass term breaks the invariance under the gauge transformation -1 1 -1 Aµ -7 s AµS.+g-S aµs where a S = exp(A (x)ta) , and the propagator for Aµ(x) once again has bad high energy behaviour. The non-abelian generalization of the Stuckelberg formulation [6] comes about through where _ U =exp(! q>(x)) =exp(! q>a(x)ta) This leads to the lagrangian 2 1 ...... 2 2 1 -1 ) ] L=-tr.t<2 -m tr[c A µ --Ug a µ U (7) 1 2 2 2 µ µ 2 = 2 tr F - m tr A + 2m tr(A aµcp) - tr(a q>Dµq>) + O(g ) (8) which is invariant under the gauge transformation 92 1 1 Aµ~ s- aµs + s- aµs u~us Once again the gauge may be fixed through the addition of L r=-2tr(s(ro(o·A+-1 B))) g 2a 2 = -2 tr(Bd·A + rod.Dro + -1 B ) (9) 2a leading to a ~ell-behaved propagator for Aµ(x). In this case though, renormalizability is not assured because of the non-polynomiality of the Lagrangian. There are an infinite number of monomial interaction terms and it is not clear whether each has to be separately renormalized. It was, however, observed [l] that the equation of motion for cp is Ocp - md.A +; g([Aµ,aµcp] + dµ[Aµ,cp]) + O(g2) = 0 or and that in the Landau gauge a ~ oo , when d· A = 0 , cp = 0 is a solution of this equation. Thus it was proposed that in the Landau gauge this could be imposed leading to the Lagrangian L =-F1 2 +-mA 1 2 + B a· A + ::;:;:".:\wu·DCO (10) 4 2 which is polynomial and gives a propagator for Aµ(x) with the appropriate high- energy behaviour. In other words, it was suggested that U or the cp field could be eliminated in (7) or (8) in favour of Aµ (x) through its equation of motion. Then in the Landau gauge the result~t lagrangian with gauge fixing would be just (10). 3. Unitarity The Lagrangian just obtained (10) does not lead to a unitary theory [2]. It may be observed that the modified BRST transformations u.r:ider which (10) is invariant s'Aµ = Dµro s'ro =B s'ro = - ~ g {ro,ro} s'B = m2ro are not nilpotent, 2 s ' ro = m2ro This in itself is perhaps not sufficient to prove that unitarity is violated. Indeed a formal argument for unitarity was given. It was based on a proof of the unitarity of the 93 original non-polynomial theory, which is invariant under a nilpotent BRST transformation, followed by manipulations of the path integral generating function so as to arrive at the theory given by the lagrangian (10). Explicit diagrammatic methods are therefore perhaps the best way to demonstrate the failure of unitarity [2],[3]. The Feyman diagram rules from (9) are: propagators PµPµ• I -Tlµµ' +-2-,- I ~ l\,/VVVV\/VVVVV A aa ,(p) = p + ie ioaa a,µ a,µ' µµ p2 - m2 + ie p ------~----- a a' vertices q,b,v p,a,µ / ~\ r,c,p g2c:::: = -ig2(fab/cd/Tlµpllvcr - llµcr'Tl vp) + facefdbe(Tlµcrll vp -Tlµv'Tlpcr) + fad/bce(Tlµv'Tlpcr -Tlµp'Tlvcr)) s,d,cr r,c,p q,b , , p,a, µ , ;'1 JVVVlfV\N.. ~ ' '...i ' 'r,c 4 4 with (21t) o (Lp.) for the incoming momenta at ·each vertex, i 1 4 d p. --~ for eai;h internal momentum, f (21t) a factor of (-1) for each ghost loop, and a symmetry factor 1/S for any symmetry of the internal lines of a graph. 1. The first process which we shall examine is the elastic scattering of two longitudinally polarized spin-1 states. For simplicity, and so that we may compare the results with those for the Higgs mechanism, we will work with an SU(2) gauge 94 ~heory, so that a= 1,2,3'' fabc = eabc. There are unitarity bounds for the scattering of two spin-1 particles [7],[8] and in order that they be satisfied the amplitude must at each order in g2 be of order 1 at high energies. Consider A1LA2L ~ A1~2L. Now at tree level, O(g2), the possible diagrams are 1 2 1 2 1 yielding contributions T5, Tt, T4 respectively to the amplitude Ttot· In the centre of mass frame Pt= (E,0,0,p), P2 = (E,0,0,-p), p1' = (E,psin8,0,pcos8), p2' = (E,-psin8,0,pcos8) with E2-p2 = m2, and the longitudinal polarization vectors are £ (p,0,0,E), = (p,0,0,-E) 1 = ~ e2 ~ e1' = ~ (p,Esin0,0,Ecos0), e2' = ~ (p,-Esin0,0,-Ecos8) with ~l = -1, p.·e.l l = 0. . We then have 2 µvp µ'v'a Ts = g E1µe2v r (Pl'P2·-P1-P2)L\pa g2 ( 2 2 2 = 4p2E (l+cos0) (4p2+2E (l-cos0)) (2p2 (l+cos8) - m 2)m 4 2 2 - 8p2E (l+cos0)(p2+E cos8)(3-cos0) 2 2 + (p2+E cos8)\4E +2p2(1-cos8))) 95 =g{U~J (3-case)(l-><:ase) + (:~.J'(- ~ -~ case)+ 0(1)) 2 ' ' µvµ'v' T 4 = g e1µ e2velµ'e2v,C = g2(U~S(cos2 e - 6cose - 3) + (!J (6cose + 2) + O(l)) and so 2 T tot = i((!) (i -i case) + 0(1)) and thus the unitarity bound is violated. By way of comparison, if masses are obtained through the Higgs mechanism, then Aµ3 is massless so that T8 = g2(4cose(!f + 0(1)) 2 Tu= g ((!~-J4(3-cos8)(1+cose) - (!r8cose + O(l)) 1 2 with T 4 as before. Further there is a coupling of the Higgs scalar to Aµ and Aµ withl/2µ>- vertex , 1/2v 1,/1giving a contribution to T tot 2),2 which is exactly wha~ is required so that Ttot should be of order 1 and so satisfy the unitarity bound. 2. Further confirmation of the violation of unitarity is given by studying the self-energy of the vector boson [2]. rf~'(p) = PµPµ' IIlong(p2) + (11µµ' - pµpµ'J IItr (p2) aa p2 aa' p2 aa' The imaginary part of IItr on shell, p2 = m2, pO>O, gives the decay width of the 96 physical spin-1 states. But if the theory does. only describe massive spin-1 bosons, then this should be zero as there are no states into which decay is possible. More importantly the negative-norm contributions from the ghosts must be fully cancelled if the theory is to be unitary. 2 Now to order g the contributions to IIµµ' are Ilvµµ', II0 µµ' and ITTµµ' given by the diagrams ,··~ ·, , ' 'VVV'V'(. hfvVv\ p~ ' ' and ' '. ·~-, ' rr~· may be broken down by splitting the vector propagator into spin-1 and spin-0 parts ~ky - I 0 'Tlµv + 2 • A;~(k) = 2 k +le ioaa k - m2 + ie -11 + ~ky ~ky µv m2 ·~aa· m2 ·~aa' = tu - lu 2 2 k - m2 + ie k + ie (l)aa' (O)aa' =Aµv (k) + Aµv (k) Then it becomes _µµ' _µµ' _µµ' _µµ' lly = 11(1)(1) + 11(0)(1) + 11(0)(0) given by (1) (1) (0) -0~··~ and -Ovvvv (1) (0) (0) The imaginary part of each contribution is given by the Cutkosky rules [8]. The tadpole diagram has no cut and so may be ignored. The imaginary part of II(l)(l) is 2 4 T _ _(_µµ' ) g d q _µvp bb'( qvqv, ) O 2 2 2llll\.ll(l)(l)aa'(p) =--;;-- J--4 i ·abc (p,q,-p-q) 8 -Tlw• +-2- 21t 8(-q) 8(q -m) L. (21t) m cc'( (p+q)p(p+q)p') 2 0 _,,pp'+ m2 21t 8(po+qo) o((p+q) -m2) _µ'v'p' (-l)l .a'b'c' (-p,-q,p+q) We need not evaluate this because here the step functions and a-functions ensure that the whole must vanish on shell, p2 = m2, p°>O. This is most easily seen in the centre of mass frame, pµ = (m,0,0,0), for then qO 97 qO~-m while pO+qO>O, (p+q)2 = m2 implies that pO+qO~m and so qo~o. and the two are not compatible. In fact complete evaluation of Im(II(l)(dP)) yields the factor 0(p0)S(p2-4m2) as we would expect for this part and in accordance with unitarity. The imaginary part or rr< 1 l;t:~~:::res on shell rr contracted wi: l 11 so as to extract the transverse part. Specifically 2m{~~·(O)aa'(p)) = g2 Jd4q4 r:: (p,q,-p-q) Bbb{- qv;v·) 21t 0(-qo)8(q2) (21t) m cc'( (p+q)P(p+q)p'J 2 . 8 -11pp~ + m2 ) 21t 8(po+qo)8((p+q) -m2) r ..Jl.'v'p' (-l)l .a'b'c'(-p,-q,p+q) 4 2 2 2 = - g f f I be f~ d (qµqp + pµqp +ppqµ + 11pµ(p - (p+q) )) abe a (21t)2 ~ 0(-qo)8(q2) S(po+ qo)8((p+q)2 - m2) m - + (p+q)p(p+q)p'J ( 11pp' 2 m I I I I I I I I 2 ( qµ qP + pµ qP + pp qµ + 1l Pµ (p2 _ (p+q) ) ) - which with p2 = m2 and the contraction with (11) understood, so that terms involving Pµ or Pµ• may be dropped , is g2 be d4q 2 = - f f , - f-- 8(-q0)8(p0+ q0)8(q2)8((p+q) - m2) m2 abe a (21t)2 p( (p+q) (p+q) ·1 I p' qµ(p+q) - 11pp' + :2 p) qµ (p+q) =0 There remains TI(O)(otµ' and TI0 µµ'. Now, with p2 = m2 and (11) understood once again 98 T .J _µµ' ) 1 2 __µ.vp bb'(- qvqv') o 2 21Ill\.ll(O)(O)aa'(p) = 2 g JA(21t)4 i ·abc (p,q,-p-q) B m2 21t 0(-q )B(q ) ·( (p+q)p(p+q) 2 Bee - m2 P·J 21t 0(po+ qo)B((p+q) ) ...Jl'v'p' (-1)1 ·a'b'c'(-p,-q,p+q) = i ::. r,,,,,r,."' J(2~) (q"(p+q)2 + (p+q/(p2 - (p+q))) 0(-qo)8(q2)0(po+ qo)8((p+q)2) µ' 2 µ' 2 ( q (p+q) + (p+q) (p2 - (p+q) ) ) and 2 2 ~ rf,i~.(p)) = - g J(~:~. ~.(p+q) a••'2x 0(-q°)8(q ) cc' o o 2 ...Jl' 8 21t 0(p + q )B((p+q) ) (-1)1 ·a'b'c'(q) 4 = - g2fabcf, be J_.9.._2 d 0(-qo)0(po+ qo)8(q2)8((p+q) 2) qµqµ ' a (21t) Thus we see that these terms are of the same sign and of the same form, but that II(O)(Otµ' only cancels half of IT0 µµ'. This factor of 112 in (0)(0) is the symmetry factor for the internal lines; of course there is no corresponding factor for the directed ghost lines. Continuuing we could confirm the violation of unitarity in other processes [3] and at higher orders but the fact of it is already clear. To understand when the violation of unitarity came about let us return to the unrenormalizable lagrangian (8)+(9). 99 4. Renormalizability n Unitarity We have seen that the renormalizable lagrangian (10), which did not possess an invariance under a nilpotent BRST transformation, does not give a unitary theory. We will examine the same processes for the unrenormalizable lagrangian (8)+(9) in the Landau gauge, 2 2 2 L = - ! tr F + i m tr A + i aµ cpDµcp +Ba.A + coa.oco + O(g ) ' which does possess a nilpotent BRST symmetry. To the Feynman rules before we must add the rules for the cp-field. p 1 ------~ iB I cp P2 + ie aa ,b,p a,µ //I \Mf\1\-( µf ', gp abc ',c together with an infinite number of other vertices with more than two attached cp-lines and a corresponding higher order of g. Then to the second process which we considered, the self energy of the vector boson, there should be added at order g2 the graph cp , , ,, ' 'VV\/Vl.i. ' '.VVV\f'v'\, ... I cp Now the Feynman rules for cp at this order are exactly those for the ghost field except that, as the lines are not directed! a symmetry factor must be inclu.ded where appropriate and, as cp is bosonic, there is no factor of (-1) for cp loops. Thus the contribution of this graph to the imaginary part of Iltr will be exactly half of that of the ghosts but with opposite sign. That is exactly what is required to restore unitarity at this order. We can see this also by examining the respective contributions of cp and the ghosts to the effective action at this order. 4 iW f ifd x L() e = N D e So the ghost term in the lagrangian coa.oro, after integrating over the complex Grassmannian ghost fields, leads to a term ln(det (a·D)) = tr(ln (a·D)) in the effective action, while the term contributes 100 1 ln(det- 2(().D)) = _ _!_ tr{ln(a.o)) 2 Presumably the terms of higher order in cp serve to remedy any other incomplete cancellations of the ghost contributions which might occur at higher orders. On the other hand the first process that we examined, the elastic scattering of two spin-1 bosons, is not altered at tree level by the reinclusion of the cp-field. The amplitude continues to violate the unitarity bound. This is because the derivation of the unitarity bound also assumes the renormalizability of the theory, and this now fails to hold. The conclusion which we must draw is that we cannot construct a consistent theory of massive, non-abelian, vector bosons without invoking the Higgs mechanism. Our attempts to ensure renormalizability were only at the expense of unitarity and vice versa. Despite this the non-abelian generalization of the Stueckelberg formulation is still interesting in the way that it demonstrates that the original naive lagrangian is of a form sufficient to ensure unitarity, and in the way in which the cancellations necessary for this unitarity are exhibited at each order. 101 References [l] R.Delbourgo and G.Thompson, Phys.Rev.Lett 57 (1986) 2610. [2] J.Kubo, Phys.Rev.Lett 58 (1987) 2000. P.Kosinski and L.Szymanowski, Phys.Rev.Lett 58 (1987) 2001. [3] R.Delbourgo, G.Thompson and S.Twisk, Int.J.Mod.Phys. A3 (1988) 435. [4] D.Boulware, AnnPhys 56 (1970) 140. [5] B.Lee, "Gauge Theories" in "Methods in Field Theory", eds R.Balain and J.Zinn-Justh?-, Les Houches (1976). [6] T.Kunimasa and T.Goto, Prog.Theor.Phys. 37 (1967) 452, [7] I.Aitchison and A.Hey, "Gauge Theories in Particle Physics," Adam Hilger (1982). [8] C.Itzykson and J.Zuber, "Quantum Field Theory," McGraw-Hill (1980). 102 Conclusion I have in this thesis considered five different topics in each of which Grassmannian variables occur in some guise. In Chapter 2 I found a square root of the Dirac equation using the superfield formulation of space-time supersymmetry. Analysis o~ the massless case showed it to be equivalent to a theory involving a massless spinor obeying the Dirac equation and a complex vector field obeying Maxwell's equations. The massive case and po:ssible interactions remain to be investigated. In the next chapter I considered the derivation of the index for the twisted Dirac operator through path integrals in a supersymmetric quantum mechanics. By considering carefully the construction of these path integrals I was able to show that the ambiguities which arise come from taking the discrete to continuum limit. Establishing a priori certain facts about the path integrals involved in the index calculation then enabled me to eliminate these ambiguities, including the overall normalization. Also, in the first part of this chapter, I used the general Atiyah-Singer index theorem result, that the index of an operator (which is elliptic, etc.) is dependent only upon the spaces between which it acts, to derive a general expression for the indices of operators between fields of arbitrary spin in the presence of a background gravitational field. This result of the index theorem that I used here has not been clearly demonstrated in the path integral derivations of the indices of individual operators as presented so far and is perhaps worthy of consideration in the future. Grand unified theories within the Grassmannian Kaluza-Klein scheme, which involves extra anticommuting coordinates rather than commuting ones, were the subject of Chapter 4. I showed that the SU(5) grand unified theory led to a more economical scheme than did the SO(lO) one, requiring only 10 extra coordinates as opposed to 64. These investigations were conducted within the framework of the Grassmannian Kaluza-Klein ansatz. For general Grassmannian Kaluza-Klein schemes it remains to be considered whether the full (4+N)-dimensional supergravity theory can lead to the ansatz in such a way that the unwanted modes are rendered harmless, and whether the 0Sp(4/N) spinor representations might be tamed. Chapter 5 was concerned with the BRST supersymmetry. It would seem that a modification of the Faddeev-Popov prescription to include the anti-BRST symmetry leads to a natural framework in which to describe the complete set of fields in a quantum theory. Further the linking of the two symmetries by an Sp(2) transformation can provide a rationaie for the occurrence of ghost fields in other than ghost-antighost pairs. I gave a formulation imposing such an Sp(2)-symmetric extended-BRST supersymmetry through a (4+2)-dimensional superspace. By not attempting to treat the full (4+2)-dimensional superspace as fundamental problems in dealing with spinors, which have arisen in other such schemes, were avoided. It would now be interesting to go on to consider whether the Sp(2) symmetry might be brought naturally and profitably into the approach of Fradkin et al. to the canonical quantization of gauge theories. Finally in Chapter 6, I discussed the renormalizability and unitarity of the 103 massive Yang-Mills theory without Higgs. I demonstrated that a scheme based on the Stueckelberg model which guarantees renormalizability does so only at the expense of unitarity. This violation of unitarity is evident perturbatively in the failure of the theory to prevent its ghost modes from appearing in the outgoing asymptotic states. The Higgs mechanism, for which such difficulties do not occur, thus emerges strengthened, though until direct experimental evidence of the Higgs is obtained investigations into possible alternatives should surely continue. 104 Awendix The purpos~ of this appendix is to explain the two component spinor notation used in Chapter 2 and to give some useful identities.. The notation is that of Wess and Bagger, except that they use m,n,etc. for vector indices while we use µ,v,etc., and their metric differs from ours by an overall sign; that is we take as the Minkowski metric (as throughout this thesis) 'Tlµv = diag(+l,-1,-1,-1) with µ,v,etc. running over 0,1,2,3. The identities given here can also be found in the appendices of their book. Undotted Greek letters at the beginning of the alphabet, cx,j3,etc., are spinor indices corresponding to the (112,0) representation of the Lorentz group, dotted ones, cx,j3,etc., are spinor indices corresponding to the conjugate representation (0,112), they all take the values 1 or 2. In accordance with spin-statistics, objects with an odd number of spinor indices usually anticommute and objects with an even number commute. The antisymmetric tensors eexP, Ecxp, eexP, Ecxp, with e12=1, e12=-l, are used to raise and lower indices from the left, i.e. ",a exp - ~j3 "' = e 'I'p , 'I'a. = ea.~ 'I'. , etc. Contracting indices gives a scalar. We take 'l'X = ~Xex = X~ex = X'I' 'l'X = 'I'a.Xa. = Xa \f = X'I' In the chiral representation a Dirac spinor is 7=(~J and the Dirac matrices are 0 Y= ( c/'), r = if·yly-f = (l OJ cf 0 0 -1 1 0 -1 OJ where a ,a2,a3 are the Pauli matrices and' a = ( , and 0 -1 ·~ ' 105 I, ,, cc?crv + CJvct)! = 211µvo! 1 (ffCJV + av~/J.. = 211µv8~ p p c1crv d' + d'av~ = 2(11µv d' + 11 vp ~ _ 11µp CJv) & CJv-&' +et CJvcf = 2(11µvet + 11vp& -11µpov) & CJvfl - et CJvcf = 2i~vpcrcra and, defining ~v = .!. ( ~v - CJvcf) 4 #V =.!.(#CJV - Q'V~)' 4 , (~VE)ap::; (cf1Ve)pa ---Jlv a~ -dJ,v ~a (er E) = (O' E) , etc. i Eµvpa CJ = CJ µv 2 pa l i ~vpcrcr = - &v ,,\/', 2 pa tr(d1v d'a) = _ .!.( µp va _ 11µa vp) _ i Eµvpcr 2 11 11 11 2 An antisymmetric two-index object must be proportional to e, so / nP = - ~ Eap'l"I' = .!.ea~'l"I', etc. rl 2 Since complex conjugation reverses the order of factors, its effect on derivatives is aa F(x,0,S) = - (-1{-.Q7F(x,0,0) a0 aea where F = 0 or 1 according as F(x,0,0) is overall commuting or anticommuting. An example illustrates this whereas a P P --0a0a m=mOa a P ~ a -~ a -P- --a-0a0 m=m8.a =--.aea 0 m=--.aea 0 m Thus, since "' ' 106 --1 107 I